A dexterous hand micro-slip feature recognition and predictive stable gripping control method

Abstract

A dexterous hand micro-slip feature recognition and predictive stable gripping control method, relying on the joint motor current and joint angle information of the dexterous hand, combining the whole machine dynamics model to solve the time sequence signal of the fingertip contact tangential force; the tangential force is decoupled into a steady-state baseline component and a high-frequency disturbance component through a time sequence trend adaptive decomposition mechanism, the noise is stripped and the weak characteristics of micro-slip initiation are highlighted; the adaptive variable weighting sliding window is used to extract the high-frequency feature energy, the baseline slope and the slope change rate multidimensional features, the multi-feature hierarchical hysteresis discrimination logic is constructed, and the micro-slip precursor is predicted in advance; a soft equal-step progressive iterative control strategy is adopted to dynamically update the normal clamping force, a maximum clamping force safety upper limit is set to avoid rigid impact and clamping shock, and when the characteristic parameters fall back to the steady-state threshold, the clamping force is locked to maintain stable gripping, and the whole process is closed loop real-time monitoring and adaptive slip suppression. The application improves the autonomous control ability of the robot dexterous hand fine work and stable holding.

Description

Technical Field

[0001] This invention belongs to the field of robot dexterous hand compliant grasping and intelligent stability control technology. Specifically, it relates to a dexterous hand weak slippage prediction and stable gripping control method based on fingertip tangential force feature decomposition. It is applicable to fine manipulation, flexible gripping and stable clamping scenarios for non-standard irregular objects, slippery objects, objects with irregular shapes and uneven mass distribution. Background Technology

[0002] As a core actuator for precise end-effector tasks in robots, the dexterous hand enables human-like flexible contact, adaptive envelope, and precise grasping operations. It holds irreplaceable value in scenarios such as unstructured environments, precise manipulation of non-standard objects, grasping flexible objects, and stable clamping of complex-shaped objects. During actual grasping and manipulation, factors such as smooth object surfaces, uneven frictional characteristics at the contact interface, irregular shapes, and offset centers of gravity can easily cause microscopic, barely perceptible slippage at the contact interface. Initially, this slippage shows no obvious positional shift, representing a latent instability state. If not identified and suppressed in time, it gradually evolves into significant posture deflection and overall slippage, directly compromising grasping stability and even causing loss of object posture or damage from impacts.

[0003] Existing dexterous hand stabilization control methods have significant limitations: First, traditional control methods relying on closed-loop positioning can only passively adjust after the object has already experienced macroscopic attitude shifts and detectable positional errors, completely lacking the ability to perceive early signs of micro-slippage, resulting in severely delayed intervention. Second, conventional gripping control only uses the overall contact force amplitude as a single threshold for judgment, without disassembling and analyzing the temporal details and fluctuation characteristics of the fingertip contact tangential force, lacking the ability to decouple steady-state trends and transient disturbances, failing to extract characteristic changes before weak slippage occurs, and exhibiting low recognition sensitivity and a lack of predictability. Third, existing anti-slip control methods mostly employ instantaneous step force application, and rigid adjustment is prone to contact impact, disrupting the flexible grip state and easily causing gripping oscillations, making it unsuitable for the precise and stable gripping of fragile, flexible, and irregularly shaped objects. Fourth, existing technologies lack feature modeling, advanced prediction, and progressive adaptive stabilization mechanisms for the evolution of weak slippage, making it difficult to adapt to the long-term stable gripping requirements of complex shapes, irregularities, and low-friction objects. Summary of the Invention

[0004] To overcome the shortcomings of existing dexterous hand gripping control methods, which can only passively detect macroscopic slippage and cannot identify microscopic weak slippage precursors in advance, and whose rigid gripping adjustment methods are prone to impact and difficult to adapt to the fine gripping of irregularly shaped, low-friction, and center-of-gravity offset objects, this invention proposes a dexterous hand weak slippage prediction and stable gripping control method based on fingertip tangential force feature decomposition. It constructs a four-level closed-loop autonomous stable gripping architecture: force feature perception, dual-domain feature fusion discrimination, compliant progressive adjustment, and steady-state self-holding maintenance. A temporal trend adaptive decomposition mechanism decouples the steady-state and disturbance components of the tangential force, and an adaptive variable weighted sliding window completes feature quantization. Multi-feature hierarchical hysteresis threshold discrimination is used to achieve microscopic slippage prediction, and a compliant progressive adaptive force adjustment strategy is matched to suppress slippage trends. This maintains stable gripping of objects for extended periods without generating rigid impacts, improving the dexterous hand's fine manipulation and compliant gripping capabilities in unstructured environments.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A method for recognizing microscopic slippage features and predictively controlling stable gripping with a dexterous hand includes the following steps: Step 1: Acquire the timing signal of the tangential force at the fingertips: After the dexterous hand completes the initial envelope gripping and initial contact positioning of the target object, a fixed sampling frequency is set to perform discrete timing sampling at equal time intervals. The timing signal of the tangential force at the contact interface between each fingertip and the object is continuously acquired in real time and recorded as follows: Simultaneously, the force signals collected from multiple fingertips are time-aligned and calibrated to unify the sampling timing reference of each signal. At the same time, the raw data streams collected in real time are buffered at fixed points and time-arranged, and a standard discrete force signal sequence is generated according to the sampling order. Step 2: Adaptive Decomposition and Decoupling of Two Components of Tangential Force Temporal Trend: A temporal trend adaptive decomposition mechanism is used to perform point-by-point baseline fitting on the original tangential force signal. The steady-state baseline of the contact force is dynamically solved through a local temporal neighborhood adaptive fitting method, replacing the traditional fixed parameter filtering method. The smooth baseline component with a slowly changing force trend is obtained adaptively. ; in, To fit the neighborhood half-window length, The adaptive attenuation weighting coefficients are dynamically fine-tuned based on the real-time fluctuation of the signal, assigning differentiated weights to force signals at different temporal locations within the neighborhood to achieve accurate fitting of steady-state trends and avoid over-smoothing of weak disturbance characteristics. Point-by-point difference calculations are performed between the original tangential force and the baseline component obtained from the adaptive fitting to decouple and separate the high-frequency fluctuation component carrying microscopic contact disturbance information. ; By decomposing the original force information into a dual-domain feature of steady-state trend component and high-frequency disturbance component, irrelevant noise interference is removed, the subtle features of the micro-slip initiation stage are highlighted, high signal-to-noise ratio feature data are obtained, and feature quantization calculation is performed. Step 3: Adaptive Variable Weighted Sliding Window Feature Calculation and Weak Slippage Multi-Feature Layered Hysteresis Joint Prediction: Abandoning the traditional fixed-length sliding window mode, an adaptive variable weighted sliding analysis window is adopted. The window's time-domain width is adaptively adjusted based on the real-time fluctuation intensity of the tangential force. Under weak disturbance conditions, the window is widened to improve steady-state smoothness, while the window is narrowed during the slippage precursor stage to improve instantaneous response capability. Simultaneously, the sliding window is controlled to shift and refresh point-by-point along the time axis with a fixed step size of a single sampling period. After each sliding operation, the latest segment of high-frequency fluctuation component data within the window is captured. The high-frequency fluctuation components of all discrete sampling points within the window are squared and accumulated to obtain the real-time window high-frequency feature energy. ; Within the same sliding window, the least squares method is used to perform linear fitting on the time-series data of the smoothed baseline component of the tangential force, and the real-time slope of the baseline component's evolution over time is obtained. Then, based on the sampling period, the rate of change of the baseline slope is further calculated. Through multiple sets of offline holding calibration tests, warning thresholds and steady-state fallback thresholds are preset respectively, and a multi-feature hierarchical hysteresis discrimination logic is constructed to distinguish between slip trigger thresholds and steady-state recovery thresholds, so as to avoid frequent false triggers caused by small feature jitters; the current feature parameters are compared with the corresponding thresholds in real time, and when multiple feature quantities exceed the warning thresholds at the same time, a slip ahead warning signal is output immediately; Step 4: Progressive Normal Clamping Force Closed-Loop Compliant Control and Steady-State Self-Supporting Clamping: Upon receiving the slippage advance warning signal, a compliant equal-step progressive anti-slip control mechanism is adopted. Following an iterative approach with fixed small force increments, the target normal clamping force command for the dexterous fingertip is recursively updated periodically. ; The maximum safe clamping force of a single fingertip is preset, and amplitude constraints are applied synchronously during each force command update to keep the normal clamping force within a safe range. The complete process of fingertip tangential force acquisition, adaptive decomposition and decoupling of temporal trends, and adaptive variable weighted sliding window feature parameter calculation is continuously and cyclically executed, and high-frequency feature energy, baseline slope, slope change rate and each preset layered hysteresis threshold are compared in real time without interruption. When multiple feature parameters fall back to within the allowable range of steady-state thresholds, the incremental superposition operation of normal clamping force is stopped, and the current clamping force value is locked to maintain a constant output. The signal acquisition and feature discrimination process is always running continuously. Once the slippage joint discrimination condition is met again during subsequent holding, the gradual iterative adjustment process of normal clamping force is immediately restarted.

[0006] The control method proposed in this invention is implemented based on a multi-degree-of-freedom bionic dexterous hand hardware platform. This dexterous hand is equipped with joint drive motors, joint position encoders, and an embedded main control unit. No force sensors or tactile sensors are installed on the fingertips. Relying on the joint motor current information and joint angle position information, the tangential force at the interface between the fingertips and the object is calculated and observed in real time through the whole machine dynamic model, providing basic input signals for subsequent micro-slip feature recognition and stable clamping control.

[0007] The beneficial effects of this invention are mainly reflected in: By performing baseline and high-frequency component dual-layer feature decomposition on the fingertip tangential force, it is possible to capture microscopic and weak slip precursors that cannot be identified by the naked eye and position detection, thereby achieving advanced prediction of instability trends and overcoming the drawback of traditional control that can only respond passively after the fact.

[0008] It adopts a smooth, step-by-step, small-increment adjustment of the normal clamping force, without rigid impact or clamping oscillation, and is suitable for the delicate and smooth gripping needs of fragile, flexible, and irregularly shaped objects.

[0009] Based on the temporal characteristics of the contact force itself, it models and judges slippage, without relying on the constraints of specific working scenarios. It is suitable for stable clamping and control of various smooth, irregular, and off-center objects in unstructured environments, and has a wide range of applications.

[0010] A closed-loop adaptive clamping mechanism is formed, which can dynamically suppress slippage in real time, greatly improving the reliability of long-term fine grip and stable operation for dexterous hands. Attached Figure Description

[0011] Figure 1 It is a flowchart of the acquisition of tangential force timing signals from dexterous fingertips and the adaptive dual-component decoupling processing of timing trends; Figure 2 This is a flowchart of adaptive variable weighted sliding window feature calculation and multi-feature hierarchical hysteresis slip joint discrimination; Figure 3 This is a flowchart of the progressive normal clamping force compliant regulation and closed-loop steady-state self-sustaining clamping control. Detailed Implementation

[0012] The invention will now be further described with reference to the accompanying drawings.

[0013] Reference Figures 1-3 A method for recognizing microscopic slippage features and predictively controlling stable gripping with a dexterous hand includes the following steps: Step 1: Establishing the dexterous hand gripping state and high-precision continuous sampling of fingertip tangential force; In this embodiment, when the dexterous hand performs a grasping task, it first completes the envelope alignment and initial gripping of the target object through the coordinated movement of the finger joints, so that each fingertip forms a stable initial contact state with the object surface. After completing the initial posture setting and contact pre-tightening, the system immediately switches to a long-term continuous monitoring and control mode. The system is pre-configured with a fixed control sampling period and a uniform sampling frequency, and performs discrete time-series sampling at equal time intervals, marking each sampling moment sequentially as a discrete sequence number. This ensures uniform timing and consistent timing reference throughout the entire force signal acquisition process. Utilizing a dexterous fingertip contact force sensing unit, the tangential interaction force at the interface between each independent fingertip and the object is synchronously, continuously, and uninterruptedly acquired, forming a one-dimensional discrete tangential force time sequence that evolves over time. This original acquired signal is defined as… In actual operation, the original tangential force signal inevitably contains various interference components such as mechanical transmission backlash jitter, micro-vibration of the connecting rod structure, low-frequency disturbances from the environment, and white noise from circuit sampling. Although the amplitude of these noises is small, they can mask the weak characteristic fluctuations in the early stages of micro-slip initiation. If directly used for feature analysis, they can easily lead to misjudgment or missed judgment of slippage. For the multi-channel force signals obtained by multi-channel synchronous fingertip acquisition, time-series alignment and calibration are performed channel by channel to unify the sampling start time and time series reference of all channels. At the same time, a dedicated data buffer area is opened to perform frame-by-frame buffering, time-series sorting and regularization of the real-time sampled data stream. The time-series aligned and sampled point-complete standardized discrete force signal sequence is generated according to the chronological order. The entire operation process of multi-channel signal synchronization, data buffering and time-series regularization is completed in sequence.

[0014] Step 2: Adaptive decomposition of tangential force temporal trend and precise decoupling of steady-state baseline and high-frequency disturbance components; In this embodiment, the original fingertip tangential force timing signal obtained in step one... The signal preprocessing is completed using a time-series trend adaptive decomposition mechanism. Through local time-series neighborhood adaptive weighted fitting, the steady-state baseline of the tangential force time-series signal is dynamically fitted. This allows for adaptive adjustment of the fitting weights based on real-time signal fluctuations, accurately distinguishing between slow-changing force trends and transient micro-perturbations. The adaptive baseline fitting calculation formula is as follows: ; in, The steady-state baseline component of the tangential force is obtained by adaptive fitting at the current sampling time. The half-window length represents the neighborhood of the local time series fitting and is used to determine the range of time series intervals covered by a single fitting. For adaptive decay weighting coefficients, The algorithm can be dynamically fine-tuned based on the real-time fluctuation amplitude of the contact force signal. Lower weights are assigned to sampling points farther from the current time, while higher weights are assigned to sampling points closer to the current time. This ensures the smoothness of the steady-state trend fitting while preserving the details of subtle fluctuations to the greatest extent possible, completely avoiding the shortcomings of traditional fixed-parameter filtering, such as smoothing out precursory features and causing phase lag. After obtaining the smoothed baseline components at each time point through adaptive fitting, the original tangential force signal at the same sampling time is compared with the baseline components point by point according to the principle of one-to-one correspondence in time sequence. This extracts the slowly changing steady-state components from the original signal and separately solves for the high-frequency fluctuation components corresponding to the micro-perturbations at the contact interface. The formula for solving the difference is as follows: ; To further suppress residual pulse interference, the obtained high-frequency fluctuation components can be subjected to mean smoothing constraint processing. A small-amplitude smoothing method using neighborhood multi-point averaging is employed, expressed as follows: ; in, The number of neighborhood half-window points, taking a small integer value, can achieve pulse spike suppression without damaging the subtle slip characteristics; the smoothed baseline component mainly reflects the long-term evolution of the object's center of gravity shifting slowly, the contact posture slightly deflecting, and the preload drifting slowly; the high-frequency fluctuation component mainly reflects the transient subtle fluctuation characteristics of the contact interface micro-friction vibration, the intermittent meshing of surface micro-protrusions, and the micro-slip initiation stage. Following the above process, the entire set of data processing operations, including original signal filtering, baseline component solving, high-frequency component separation, and high-frequency component secondary smoothing, are completed in sequence.

[0015] Step 3: Adaptive variable weighted sliding window feature calculation, baseline temporal slope fitting, and multi-feature hierarchical hysteresis weak slip joint prediction; In this embodiment, the traditional time-series sliding window with a fixed number of sampling points is abandoned, and an adaptive variable weighted sliding analysis window is adopted. The number of sampling points accommodated by the window is adaptively adjusted in real time according to the amplitude of the high-frequency fluctuation of the tangential force. The system updates point-by-point along the time axis with a single sampling period as the step size, always maintaining the latest continuous time-series data within the window to ensure the real-time performance and environmental adaptability of feature analysis. Within the current sliding window range, the preprocessed high-frequency fluctuation components are calculated using time-domain energy accumulation. The characteristic amplitude of weak, small-amplitude fluctuations is amplified by square accumulation, constructing a high-frequency characteristic energy index that can quantify the intensity of micro-tremors at the contact interface. The calculation formula is as follows: ; To mitigate the impact of abrupt changes at the window boundary, a Hanning window weighted smoothing process can be used. The weighted energy expression is as follows: ; in, Weighting coefficients for the Hanning window, and applying normalization constraints to the window function coefficients: This is used to suppress spectral leakage and improve the stability of characteristic energy calculation. Simultaneously, within the same sliding window, a univariate linear regression is performed using the least squares method to fit the smoothed baseline component sequence of the tangential force, solving for the instantaneous slope of the baseline component over time. This slope characterizes the slow-varying drift trend of the overall clamping force. The least squares slope fitting formula is: ; To characterize the drastic change in the slip trend, the slope change rate parameter is further defined: ; in, The system sampling period is used to characterize the rate of baseline drift and enrich the dimensions of slippage discrimination features. Multiple offline grip calibration experiments were conducted, selecting target objects with different surface materials and shapes for repeated clamping tests. Multiple sets of characteristic parameter values ​​under steady-state and micro-slippage states were recorded. Based on the experimental statistical results, warning trigger thresholds and steady-state fallback thresholds were set for high-frequency characteristic energy, baseline slope, and slope change rate. A multi-feature hierarchical hysteresis discrimination mechanism was constructed, and a constraint relationship was set between the two sets of thresholds. ; in, and These are the early warning trigger threshold and steady-state fallback threshold in the high-frequency characteristic energy dimension, respectively, used to determine whether the intensity of high-frequency micro-vibration disturbance at the contact interface reaches the level of micro-slip precursor. and These are the early warning trigger threshold and steady-state fallback threshold, respectively, for determining whether the slow-changing drift trend of the clamping force conforms to the slip evolution characteristics. and These are the warning trigger threshold and steady-state fallback threshold, representing the rate of change of the baseline slope, respectively, used to characterize the severity of the slippage trend. The system determines that a slippage warning requires all three features to exceed their respective warning trigger thresholds simultaneously, and the warning is lifted when all three features fall back to within their respective steady-state fallback thresholds. This constitutes a multi-dimensional joint hysteresis discrimination logic.

[0016] The system compares each calculated feature parameter with its corresponding threshold in real time. When all feature values ​​exceed their respective warning trigger thresholds simultaneously, the system internally marks the status and immediately triggers a slippage advance warning signal. The warning trigger condition is as follows: ; in, The characteristic energy of the high-frequency fluctuation component of the tangential force within the sliding window at the k-th sampling time is used to quantify the severity of micro-tremors at the contact interface. It is the absolute value of the slope of the steady-state baseline fitting of the tangential force during the same period, used to characterize the slow drift trend of the clamping force; It is the absolute value of the rate of change of the baseline slope during the same period, used to characterize the rate of evolution of the slip trend.

[0017] When the high-frequency characteristic energy, baseline slope, and rate of change of slope simultaneously fall back to within their respective steady-state fallback thresholds during the subsequent holding period, the system lifts the warning status. The steady-state fallback condition is as follows: ; The parameters are consistent with those in the above formula, so their meanings will not be repeated here.

[0018] Step 4: Iterative control of progressive normal clamping force, determination of steady-state constraints, and adaptive closed-loop clamping maintenance; In this embodiment, after the system outputs a slippage warning signal, it abandons the rigid adjustment method of increasing the clamping force by a step abruptly and instantaneously in the traditional dexterous hand control, which avoids damage to the object surface, clamping posture oscillation, and sudden changes in contact stress caused by instantaneous contact impact. Instead, it adopts a compliant, step-by-step, gradual slippage control mechanism. With a small increment, continuous iteration, and smooth gradual control strategy, it iteratively updates the target normal clamping force command value of the dexterous fingertip on a sampling cycle. The clamping force iterative update formula is as follows: ; in, This is the normal clamping force control command output at the current sampling moment. This represents the steady-state value of the normal clamping force corresponding to the steady-state grip at the previous sampling time. To preset a constant, small force increment, a reasonable amplitude constraint is set for the force increment: ; The force values ​​are kept small and gentle to ensure that each force adjustment is smooth and without sudden changes or rigid impacts. To prevent the clamping force from accumulating without limit, an upper limit constraint is set on the normal clamping force: ; in, To ensure maximum safe clamping force per fingertip and prevent over-clamping from causing object deformation or structural overload, the system continuously performs closed-loop cyclic operation of the aforementioned tangential force sampling, adaptive decomposition and decoupling of temporal trends, and adaptive variable weighted sliding window feature calculation. It updates high-frequency feature energy, baseline slope, and slope change rate parameters in real time, dynamically monitoring the attenuation and evolution of slip characteristics at the contact interface. As the normal clamping force steadily and slightly increases, the normal pressure at the contact interface increases, the maximum static friction constraint capability of the interface simultaneously strengthens, and the microscopic slip precursor characteristics gradually attenuate and decline. When all real-time feature parameters synchronously return to the steady-state threshold safe range, the steady-state clamping constraint condition is met. ; At this point, it is determined that the micro-slippage trend at the contact interface has been completely suppressed, and the object's clamping posture has returned to a stable steady state. The incremental iterative superposition of the normal clamping force is immediately stopped, and the current normal clamping force command value is locked to maintain a constant output, ensuring stable holding of the object for a long time. At the same time, the system does not terminate the signal acquisition and feature monitoring process, but continues to perform time-series sampling, component decoupling, and feature discrimination without interruption. If, during subsequent holding, factors such as slight external disturbances, attenuation of the friction characteristics of the contact surface, or slight changes in the object's own posture meet the slippage joint prediction conditions again, the system will automatically restart the progressive normal clamping force iterative control process, repeating the closed-loop adaptive slippage suppression, enabling the dexterous hand to have complete intelligent control capabilities, including autonomously sensing weak slippage precursors, making advance predictions, flexibly and progressively adjusting force, and maintaining self-stability.

[0019] The method in this embodiment is implemented based on a multi-degree-of-freedom bionic dexterous hand hardware platform. The dexterous hand is equipped with joint drive motors, joint position encoders and embedded main control units, and there is no need to install force sensors or tactile sensors on the fingertips. The tangential force of the fingertip contact interface is obtained in real time by solving and observing the joint motor current information and joint angle position information through the whole machine dynamic model.

[0020] The embodiments described in this specification are merely illustrative examples of specific implementations of the technical concept of the present invention. The scope of protection of the present invention should not be limited to the specific technical solutions disclosed in these embodiments, but also covers all equivalent technical means that can be obtained by those skilled in the art through conventional deduction, equivalent substitution, or simple modification based on the overall concept of the present invention.

Claims

1. A method for recognizing and predictively controlling the stable gripping of microscopic slippage features of a dexterous hand, characterized in that, The method includes the following steps: Step 1: Acquisition and Timing Warping of Tactile Force Signals at Fingertips: After the dexterous hand completes the initial envelope gripping and initial contact positioning of the target object, a fixed sampling frequency is set to perform discrete timing sampling at equal time intervals, continuously acquiring the tangential force timing signals at the contact interface between each fingertip and the object in real time. The timing alignment and calibration of multiple fingertip force signals are performed to unify the sampling timing reference, and the original data stream is buffered and time-arranged to generate a standard discrete force signal sequence. Step 2, Adaptive Decomposition of Temporal Trend of Tangential Force and Decoupling of Two Components: An adaptive decomposition mechanism of temporal trend is adopted, and the steady-state baseline component of tangential force is solved by adaptive weighted fitting of local temporal neighborhood. The original tangential force and the steady-state baseline component are subtracted point by point to decouple and separate the high-frequency fluctuation component that carries the information of microscopic contact disturbance. ; Step 3: Adaptive Variable Weighted Sliding Window Feature Calculation and Weak Slippage Multi-Feature Hierarchical Hysteresis Joint Prediction: An adaptive variable weighted sliding window is used, which adaptively adjusts the time-domain width of the window according to the real-time fluctuation intensity of the tangential force, and refreshes the window point by point with a fixed single sampling period step size; the real-time high-frequency feature energy of the high-frequency fluctuation components within the window is calculated. The slope of the real-time baseline change was obtained by linear fitting of the steady-state baseline components using the least squares method. Further calculate the rate of change of the baseline slope. By setting warning thresholds and steady-state fallback thresholds through offline holding calibration tests, a multi-feature hierarchical hysteresis discrimination logic is constructed. When multiple feature quantities, such as high-frequency feature energy, baseline slope, and slope change rate, exceed the warning thresholds simultaneously, a slippage advance warning signal is output. Step 4: Gradual Normal Clamping Force Closed-Loop Compliant Control and Steady-State Self-Supporting Clamping: Upon receiving a slippage advance warning signal, a compliant equal-step gradual anti-slip control mechanism is adopted to iteratively update the target normal clamping force command of the dexterous fingertip on a sampling cycle. The maximum safe clamping force of a single fingertip is set as an amplitude constraint; the process of tangential force acquisition, component decoupling, and characteristic parameter calculation is executed in a loop, and the characteristic parameters are compared with the steady-state threshold in real time; when multiple characteristic parameters fall back to the steady-state threshold range simultaneously, the superposition of normal clamping force increment is stopped, and the current clamping force is locked to a constant output; continuous monitoring and discrimination are performed, and when the slippage discrimination condition is met again, the gradual control is restarted to form a closed-loop adaptive stable clamping control.

2. The method for identifying and predictively gripping stable objects using microscopic slippage features of a dexterous hand as described in claim 1, characterized in that... In step 2, the formula for calculating the steady-state baseline component of the tangential force is: ； in, L To fit the neighborhood half-window length, The adaptive attenuation weighting coefficient is dynamically fine-tuned based on the real-time fluctuation of the signal, and its value satisfies... ; The formula for calculating the high-frequency fluctuation component is: ； The high-frequency fluctuation components are subjected to a secondary smoothing constraint by averaging multiple points in the neighborhood. The expression is as follows: ； In the formula, N The number of points in the neighborhood half-window.

3. The method for identifying and predictively gripping stable objects using microscopic slippage features of a dexterous hand as described in claim 1 or 2, characterized in that... In step 3, the high-frequency feature energy of the real-time window is calculated using Hanning window weighted smoothing, and the formula is as follows: ； in, To adapt the number of sampling points in the sliding window, The weighting coefficients for the Hanning window satisfy the normalization constraint. .

4. A method for identifying and predictively gripping stable objects using microscopic slippage features of a dexterous hand as described in claim 1 or 2, characterized in that... In step 3, the least squares fitting formula for the real-time slope of the baseline is: ； The formula for calculating the rate of change of the baseline slope is: ； in, This is the system sampling period.

5. A method for identifying and predictively gripping stable objects using microscopic slippage features of a dexterous hand as described in claim 1 or 2, characterized in that... In step 3, the multi-feature hierarchical hysteresis threshold is constructed to satisfy the size constraint: ； in, The high-frequency characteristic energy early warning trigger threshold, The high-frequency characteristic energy steady-state fall-off threshold, The threshold for triggering a warning of the steady-state baseline slope of the tangential force is set as follows. The steady-state baseline slope of the tangential force is the steady-state fall-off threshold. The baseline slope change rate warning trigger threshold, The steady-state fall-off threshold for the rate of change of the baseline slope; The conditions for triggering a slippage warning are: ； in, The characteristic energy of the high-frequency fluctuation component of the tangential force at the k-th sampling time is... Let be the absolute value of the slope of the steady-state baseline fitting of the tangential force at the k-th sampling time. The absolute value of the rate of change of the baseline slope at the k-th sampling time; The steady-state clamping constraint condition is: 。 6. A method for identifying and predictively gripping stable objects using microscopic slippage features of a dexterous hand as described in claim 1 or 2, characterized in that... In step 4, the formula for iteratively updating the normal clamping force command is: ； in, This represents the steady-state gripping normal force value at the previous sampling time. To preset a constant, small force increment, satisfying ; The constraint condition for the amplitude of the normal clamping force is: ； in, The maximum safe clamping force allowed for a single fingertip.

Images