A multi-dimensional tactile sensing method for a robotic fingertip

Abstract

The application discloses a kind of robot fingertip multidimensional tactile sensing methods, comprising the following steps: step S1: the multichannel strain signal transmitted by flexible cover layer to the rigid support structure is collected, and original sensing data is obtained;Step S2: the three-dimensional component of external contact force and equivalent contact point position are calculated by force and torque conservation principle, and the touch estimation result based on mechanics is obtained;Step S3: nonlinear mapping is carried out by machine learning model, and the three-dimensional component of external contact force and contact position parameter are directly output, and the touch estimation result based on data driving is obtained;Step S4: dynamic modeling and compensation are carried out on the dynamic effect caused by flexible cover layer, and the fused tactile perception result is obtained;Step S5: the fused tactile perception result is output, and the fused tactile perception result includes three-dimensional contact force size and direction and contact position coordinate.The application can make tactile information be used for robot gripping, operation and closed-loop control stably.

Description

Technical Field

[0001] This invention relates to the field of robot perception and intelligent manufacturing technology, specifically to a multi-dimensional tactile sensing method for robot fingertips. Background Technology

[0002] With the widespread application of robots in industrial manufacturing, service robots, and human-robot collaboration, the robot's ability to perceive its environment has become a crucial factor restricting its operational dexterity and intelligence level. Among these, the robot's end effector (especially the fingertip) comes into direct contact with external objects during grasping, manipulation, and interaction. Its ability to perceive tactile information such as the magnitude, direction, and position of contact force plays a key role in achieving stable grasping, slip detection, posture adjustment, and precise manipulation.

[0003] Existing robot tactile sensing technologies mainly include the following categories: One type is force / torque sensors based on rigid structures. These solutions typically employ a metallic elastomer combined with strain gauges, piezoelectric or capacitive measurement structures, enabling relatively accurate measurement of three-dimensional or six-dimensional force / torque information. They offer advantages such as high linearity, fast response speed, and good long-term stability. However, these sensors have relatively high structural rigidity, and their size and thickness are difficult to compress, making them unsuitable for space-constrained areas such as robot fingertips. Furthermore, their ability to sense contact position and local contact state is limited, making it difficult to meet the fine tactile requirements of complex contact scenarios.

[0004] Another type is tactile sensing based on flexible arrays, such as capacitive, piezoresistive, or piezoelectric flexible tactile arrays. These schemes achieve high spatial resolution by integrating a large number of sensing units on a flexible substrate, reflecting the pressure distribution within the contact area, and have been widely studied in the fields of electronic skin and flexible tactile sensing. However, due to the significant viscoelastic properties of flexible materials, these sensors are prone to hysteresis, drift, and nonlinear errors during dynamic loading and unloading, resulting in insufficient real-time performance and stability of force values ​​and contact positions. Furthermore, multi-channel array structures also suffer from severe signal coupling, complex calibration, and high manufacturing costs, making them unsuitable for long-term applications in high-reliability scenarios such as robotic fingertips.

[0005] In addition, in recent years, tactile sensing solutions based on vision or optics have emerged, which use cameras to observe the internal or surface deformation of flexible media to deduce contact force and contact location. These solutions have certain advantages in spatial resolution, but their system structure is relatively complex, they are highly dependent on lighting, camera calibration, and computing resources, and their real-time performance and robustness are still insufficient in high-speed dynamic interaction and closed-loop control scenarios.

[0006] To balance structural stability and surface compliance, some studies have proposed a rigid-flexible hybrid tactile sensing structure. This involves arranging a limited number of force or strain sensing units beneath a flexible overlay, inferring external contact information through structural force transmission relationships. While this approach reduces the number of sensing channels and improves structural reliability to some extent, current technologies primarily focus on estimating resultant forces or simple contact states. They lack sufficient accuracy for decoupling and reconstructing three-dimensional forces and contact positions at the robot's fingertip scale. Furthermore, the viscoelastic hysteresis and slowly varying bias introduced by the flexible overlay are typically addressed using simple filtering or static compensation methods, making it difficult to simultaneously balance response speed and measurement stability during dynamic operations.

[0007] In summary, existing robot fingertip tactile sensing technologies generally suffer from the following shortcomings: (1) It is difficult to achieve high-precision sensing of three-dimensional force and contact position simultaneously in a compact fingertip structure; (2) The viscoelastic effect introduced by flexible materials leads to slow dynamic response, large hysteresis, and insufficient long-term stability; (3) The number of sensing structures or channels is large, resulting in high system complexity, which is not conducive to engineering applications and large-scale deployment; (4) There is a lack of highly reliable tactile sensing methods that are deeply coupled with robot closed-loop control. Summary of the Invention

[0008] To overcome the shortcomings of the prior art, the present invention aims to provide a multi-dimensional tactile sensing method for robot fingertips, which can achieve synchronous, high-precision, and low-hysteresis sensing of three-dimensional contact force and contact position during robot fingertip contact under the conditions of compact sensing structure and limited number of sensing channels, and effectively suppress dynamic hysteresis, slow-changing bias and measurement noise caused by flexible covering layer, so that tactile information can be stably used for robot grasping, operation and closed-loop control.

[0009] To achieve the objective of this invention, the following solution is adopted: A multi-dimensional tactile sensing method for robot fingertips, applied to a robot fingertip tactile sensing device with a rigid-flexible hybrid structure, includes the following steps: Step S1: During the contact between the robot's fingertip and an external object, multiple multi-axis strain sensing units set on the rigid support structure are used to collect multi-channel strain signals transmitted from the flexible covering layer to the rigid support structure to obtain raw sensing data. Step S2: Using the original sensing data and based on the spatial geometric layout of the rigid support structure, calculate the three-dimensional components of the external contact force and the equivalent contact point position through the principle of conservation of force and torque, and obtain the tactile estimation result based on mechanics. Step S3: Using the original sensing data, perform nonlinear mapping through a machine learning model to directly output the three-dimensional components of the external contact force and the contact position parameters, and obtain the data-driven tactile estimation result; Step S4: The mechanical tactile estimation result and the data-driven tactile estimation result are used as multi-source observation information. State estimation and filtering methods are introduced for fusion calculation. Dynamic modeling and compensation are performed on the dynamic effects caused by the flexible covering layer to obtain the fused tactile perception result. Step S5: Output the fused tactile perception result, which includes the magnitude and direction of the three-dimensional contact force and the coordinates of the contact position.

[0010] Furthermore, the state estimation and filtering method in step S4 is specifically the Kalman filtering method, which recursively estimates the tactile state by constructing a state space model.

[0011] Furthermore, the machine learning model in step S3 is a multilayer perceptron model, which is used to realize the nonlinear end-to-end mapping from multi-channel raw tactile data to three-dimensional contact force and contact position.

[0012] Furthermore, the principle of force and torque conservation in step S2 specifically includes: based on the principle of force balance and the mapping method of the center of pressure force application point, the multi-channel strain signal is analyzed and calculated to obtain the three-dimensional force components and contact position.

[0013] Furthermore, the robotic fingertip tactile sensing device with the rigid-flexible hybrid structure includes: a flexible covering layer located on the outermost layer for direct contact with objects, a rigid support structure fixedly connected to the flexible covering layer, and a plurality of multi-axis strain sensing units arranged on the rigid support structure.

[0014] Furthermore, the three-dimensional contact force magnitude and direction, as well as the contact position coordinates output in step S5, are directly used for robot gripping force adjustment or posture control, forming a closed-loop control driven by tactile feedback.

[0015] Furthermore, the dynamic effects in step S4 include low-pass hysteresis caused by the flexible overlay, slow-varying bias, and random noise; the dynamic modeling and compensation are used to suppress the above effects and improve the temporal consistency and long-term stability of the tactile signal.

[0016] Furthermore, in step S4, a state-space modeling approach is adopted, in which the external contact force and contact position are regarded as system states that change over time, and the system states are recursively estimated in a continuous time series.

[0017] Furthermore, the machine learning model used in step S3 is pre-trained using a dataset and is used to compensate for deviations in the results of the physical perception calculation steps at the edge of the force-bearing area and within a specific force amplitude range.

[0018] Furthermore, the three-dimensional components of the external contact force calculated in step S2 include the normal force and the tangential force; the contact position coordinates are the two-dimensional coordinates of the contact point on the plane where the rigid support structure is located.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention features a compact structure and is suitable for fingertip integration. The invention employs a rigid-flexible hybrid tactile sensing device. By arranging a limited number of multi-axis strain sensing units on a rigid support structure to collect signals, it eliminates the need for large-scale sensor arrays, thereby achieving compact and easily integrated multi-dimensional tactile perception in space-constrained scenarios such as robot fingertips.

[0020] 2. This invention can simultaneously acquire three-dimensional force and contact position. On the one hand, this invention calculates three-dimensional force and contact position based on mechanical constraints; on the other hand, it directly outputs three-dimensional force and contact position based on a data-driven model. After fusing the two, the final output includes complete tactile information containing the magnitude and direction of the three-dimensional contact force and the coordinates of the contact position. This achieves synchronous perception of three-dimensional force and contact position without increasing the number of sensing channels, and can meet the actual needs of precise grasping and manipulation.

[0021] 3. This invention can effectively suppress hysteresis and drift caused by flexible materials. This invention introduces state estimation and filtering methods to dynamically model and compensate for the dynamic hysteresis, slow-varying bias, and random noise caused by the viscoelasticity of the flexible covering layer. This significantly reduces the hysteresis and measurement drift caused by flexible materials, thereby improving the temporal consistency and long-term stability of the tactile signal.

[0022] 4. This invention is suitable for closed-loop control and engineering applications. The fused tactile perception results output by this invention have good real-time performance and stability, and can be directly used for robot grasping force adjustment, posture optimization, and interactive control of tactile actuation, effectively improving the safety and reliability of robot operation and meeting the requirements of robot closed-loop control for highly reliable tactile perception.

[0023] 5. This invention reduces system complexity and manufacturing costs. This invention requires only a limited number of strain sensing units and, by combining a fusion strategy of mechanical models and data-driven models, significantly reduces the number of sensing units and calibration complexity. Compared to high-density flexible arrays or visual tactile solutions, it significantly reduces system complexity and manufacturing costs, which is beneficial for system engineering implementation and large-scale application. Attached Figure Description

[0024] Figure 1 This is a flowchart of a robot fingertip multidimensional tactile sensing method in an embodiment of the present invention. Detailed Implementation

[0025] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0026] This invention provides a multi-dimensional tactile sensing method for robot fingertips. Through a rigid-flexible coupling structure and strain sensing, it achieves the synchronous acquisition and reconstruction of three-dimensional force information and contact position during robot fingertip contact. This method can be used for tactile feedback in robot grasping, manipulation, and closed-loop control. This multi-dimensional tactile sensing method for robot fingertips is applicable to fingertip tactile perception and interactive control in industrial robots, service robots, and dexterous manipulation robots.

[0027] like Figure 1 As shown, the robot fingertip multidimensional tactile sensing method of this invention is applied to a robot fingertip tactile sensing device with a rigid-flexible hybrid structure. The method includes the following steps: Step S1: During the contact between the robot's fingertip and an external object, multiple multi-axis strain sensing units set on the rigid support structure collect multi-channel strain signals transmitted from the flexible covering layer to the rigid support structure to obtain raw sensing data.

[0028] In this embodiment, step S1 is the multi-channel tactile signal acquisition step, the details of which are as follows: During the robot's grasping or manipulating tasks, when the fingertip comes into contact with an external object, the flexible cover layer undergoes elastic deformation. The external contact force is transmitted to the rigid support structure through the flexible cover layer, and measurable strain or force signals are generated at multiple strain sensing units.

[0029] The output signals of each strain sensing unit are collected synchronously by the data acquisition module to form multi-channel raw tactile data.

[0030] Step S2: Using the original sensing data and based on the spatial geometric layout of the rigid support structure, calculate the three-dimensional components of the external contact force and the equivalent contact point position through the principle of force and torque conservation, and obtain the tactile estimation result based on mechanics.

[0031] In this embodiment, step S2 is the tactile information calculation step based on the mechanical model, and the specific content is as follows: Based on the spatial distribution relationship of each strain sensing unit in the rigid support structure, a force balance model between the external contact force and the output of each strain sensing unit is established.

[0032] Based on the principle of conservation of force and torque, the multi-channel tactile signals are calculated to obtain the three-dimensional components of the external contact force, and the position parameters of the equivalent contact point on the rigid support structure are further calculated.

[0033] Step S3: Using the original sensing data, a machine learning model is used to perform nonlinear mapping, directly outputting the three-dimensional components of the external contact force and the contact position parameters, thus obtaining a data-driven tactile estimation result. Step S3 can compensate for the nonlinear effects caused by complex contact conditions and non-ideal structures.

[0034] In this embodiment, step S3 is a tactile information estimation step based on a data-driven model, the specific content of which is as follows: A data-driven model is constructed, which takes multi-channel raw tactile data as input and three-dimensional contact force and contact position as output, to model complex nonlinear mapping relationships.

[0035] The multi-channel tactile signals collected in step S1 are input into the data-driven model to obtain another set of estimation results for external contact force and contact position.

[0036] Step S4: The mechanically based tactile estimation result and the data-driven tactile estimation result are used as multi-source observation information. State estimation and filtering methods are introduced for fusion calculation, and dynamic modeling and compensation are performed on the dynamic effects caused by the flexible covering layer to obtain the fused tactile perception result.

[0037] In this embodiment, step S4 is the multi-source tactile information fusion and dynamic compensation step, the specific content of which is as follows: The tactile estimation results based on the mechanical model obtained in step S2 and the tactile estimation results based on the data-driven model obtained in step S3 are used as multi-source observation information and fused by the state estimation method.

[0038] During the fusion process, the dynamic hysteresis effect, slow-varying bias and random noise introduced by the flexible overlay layer are uniformly modeled and compensated to obtain temporally continuous, stable and reliable tactile perception results.

[0039] Step S5: Output the fused tactile perception result, which includes the magnitude and direction of the three-dimensional contact force and the coordinates of the contact position.

[0040] In this embodiment, step S5 is the tactile information output and application step, the specific content of which is as follows: The system outputs fused and compensated tactile information, including the magnitude and direction of three-dimensional contact force and the coordinates of the contact position, for robot grasping force adjustment, posture control, or tactile-driven operation tasks.

[0041] In some embodiments, the multi-axis strain sensing unit is a three-dimensional force sensing unit based on the strain principle, and the flexible covering layer is an elastomer material used to improve the fit and contact stability between the fingertip and the object surface.

[0042] In some embodiments, step S4 uses state-space modeling to recursively estimate the tactile state, so as to simultaneously take into account the tactile response speed and long-term stability.

[0043] The following is a further detailed description of the robot fingertip multidimensional tactile sensing method according to an embodiment of the present invention.

[0044] The multi-dimensional tactile sensing method for robot fingertips in this invention is applied to the fingertips of a robot end effector to obtain three-dimensional contact force and contact position in real time during the contact between the robot and external objects.

[0045] The specific details of the robot fingertip multidimensional tactile sensing method according to this invention are as follows: In this embodiment, the construction of the tactile sensing structure is detailed as follows: A rigid-flexible hybrid robotic fingertip tactile sensing structure is constructed, comprising a flexible covering layer, a rigid support structure, and multiple multi-axis strain sensing units disposed on the rigid support structure.

[0046] The flexible covering layer is located on the outermost layer of the fingertip and is used to make direct contact with external objects to improve the contact fit and the continuity of force transmission. The rigid support structure is located below the flexible covering layer and is used to bear external loads and provide a stable mechanical reference. Multiple multi-axis strain sensing units are distributed in different positions of the rigid support structure to sense structural deformation or force changes caused by external contact.

[0047] The following explains dynamic compensation based on multi-source information fusion: In this embodiment, the haptic information fusion steps are further explained to address the hysteresis and drift problems introduced by the flexible cover material during dynamic loading and unloading.

[0048] In the tactile information fusion process, external contact force and contact position are regarded as system states that change over time, and a corresponding state-space model is established. By recursively estimating the tactile state, the system can suppress measurement noise in continuous time series and compensate for low-pass hysteresis and slow-varying bias caused by flexible materials.

[0049] To achieve accurate estimation of concentrated external forces on soft surfaces while eliminating adverse effects such as viscoelastic hysteresis introduced by the soft material, this paper proposes a multi-source information fusion method. This method first utilizes two independent signal observation paths to calculate external force information from the three-dimensional force readings of four three-dimensional force sensing units. The first observation path uses analytical calculation based on the force balance principle and the COP force application point mapping method. This method yields good continuity and is sensitive to tangential forces, but its overall accuracy is low, and it introduces additional noise during the COP conversion process. The second observation layer employs an end-to-end mapping based on deep learning, directly outputting the three-dimensional components of the external force and the coordinates of its application point. However, the accuracy of this method is limited by the training dataset, and deviations may occur at the edges of the force-bearing region and within specific force amplitude ranges.

[0050] Due to the viscoelastic properties of soft materials, the observation data from both methods contain a first-order low-pass hysteresis effect introduced by the force transmission path and signal readout circuit, and each channel has independent, slowly varying measurement bias and process noise. To comprehensively utilize the advantages of both methods and suppress noise and systematic errors, this paper introduces a two-stage Kalman filter for data fusion and designs a multilayer sensing and filtering fusion algorithm framework. This framework deeply couples multilayer perceptron (MLP) regression, center of pressure (COP) reconstruction, and Kalman filtering (KF) to obtain high-fidelity tactile estimation output. This can then be used to construct a closed-loop control strategy to optimize the dexterous operation of the robotic arm.

[0051] This method improves the dynamic response speed and long-term stability of tactile perception results without increasing the number of sensing units, making it more suitable for robot closed-loop control applications.

[0052] The following explains the application of robot grasping and manipulation: In this embodiment, the robot fingertip multidimensional tactile sensing method is applied to robot grasping and manipulation scenarios.

[0053] During the grasping process, the robot adjusts the grasping force applied by its fingertips based on real-time output of three-dimensional contact force information to prevent object slippage or damage. Simultaneously, it uses information on changes in contact position to correct the grasping posture, improving grasping stability. The robotic arm grasps objects made of different materials (ping-pong balls, tennis balls, and disposable plastic cups), which vary significantly in mass and stiffness. Sensors measure the contact force distribution in real time, and the robotic arm optimization controller calculates the optimal fingertip force, thereby preventing slippage and object damage during grasping.

[0054] Within a closed-loop control framework based on tactile feedback, the controller estimates the current gripping posture by analyzing the six-dimensional contact force and force point coordinates provided by the sensors in real time, and then solves for the optimal gripping posture. Subsequently, the system uses feedback control to adjust the output of each joint of the robotic arm, ensuring that the line connecting the object's center of gravity and the gripping point remains perpendicular to the ground. In this posture, only pure external forces exist between the object and the sensors, without any external torques, thus achieving optimal force alignment.

[0055] In the human tactile system, the central nervous system integrates tactile cues such as fingertip force and estimates states over time, forming an internal representation of the contact process and supporting continuous inference of contact position, relative motion, and interaction pattern evolution. Inspired by this, we designed a corresponding control and inference module to perform spatiotemporal recursive estimation of contact states driven by tactile signals. This allows robots equipped with tactile sensors to reconstruct the dynamic evolution of the contact process (including contact position trajectory, sliding / rolling, etc.) solely based on tactile information in their "internal representation." Through this design, the robot can not only adjust force through active contact but also infer the temporal trajectory of its fingertips based on tactile feedback generated by passive motion, thereby significantly improving its ability to perceive and interpret external contact events.

[0056] During operation, by continuously acquiring the temporal changes in contact force and contact position, the robot can perceive and judge the contact process, enabling it to complete fine manipulation tasks based on tactile feedback.

[0057] The robot fingertip multidimensional tactile sensing method of this invention has the following advantages: 1. This invention features a compact structure and is suitable for fingertip integration. The invention employs a rigid-flexible hybrid tactile sensing device. By arranging a limited number of multi-axis strain sensing units on a rigid support structure to collect signals, it eliminates the need for large-scale sensor arrays, thereby achieving compact and easily integrated multi-dimensional tactile perception in space-constrained scenarios such as robot fingertips.

[0058] 2. This invention can simultaneously acquire three-dimensional force and contact position. On the one hand, this invention calculates three-dimensional force and contact position based on mechanical constraints; on the other hand, it directly outputs three-dimensional force and contact position based on a data-driven model. After fusing the two, the final output includes complete tactile information containing the magnitude and direction of the three-dimensional contact force and the coordinates of the contact position. This achieves synchronous perception of three-dimensional force and contact position without increasing the number of sensing channels, and can meet the actual needs of precise grasping and manipulation.

[0059] 3. This invention can effectively suppress hysteresis and drift caused by flexible materials. This invention introduces state estimation and filtering methods to dynamically model and compensate for the dynamic hysteresis, slow-varying bias, and random noise caused by the viscoelasticity of the flexible covering layer. This significantly reduces the hysteresis and measurement drift caused by flexible materials, thereby improving the temporal consistency and long-term stability of the tactile signal.

[0060] 4. This invention is suitable for closed-loop control and engineering applications. The fused tactile perception results output by this invention have good real-time performance and stability, and can be directly used for robot grasping force adjustment, posture optimization, and interactive control of tactile actuation, effectively improving the safety and reliability of robot operation and meeting the requirements of robot closed-loop control for highly reliable tactile perception.

[0061] 5. This invention reduces system complexity and manufacturing costs. This invention requires only a limited number of strain sensing units and, by combining a fusion strategy of mechanical models and data-driven models, significantly reduces the number of sensing units and calibration complexity. Compared to high-density flexible arrays or visual tactile solutions, it significantly reduces system complexity and manufacturing costs, which is beneficial for system engineering implementation and large-scale application.

[0062] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for multidimensional tactile sensing at the fingertips of a robot, characterized in that, The method, applied to a robotic fingertip tactile sensing device with a rigid-flexible hybrid structure, includes the following steps: Step S1: During the contact between the robot's fingertip and an external object, multiple multi-axis strain sensing units set on the rigid support structure are used to collect multi-channel strain signals transmitted from the flexible covering layer to the rigid support structure to obtain raw sensing data. Step S2: Using the original sensing data and based on the spatial geometric layout of the rigid support structure, calculate the three-dimensional components of the external contact force and the equivalent contact point position through the principle of conservation of force and torque, and obtain the tactile estimation result based on mechanics. Step S3: Using the original sensing data, perform nonlinear mapping through a machine learning model to directly output the three-dimensional components of the external contact force and the contact position parameters, and obtain the data-driven tactile estimation result; Step S4: The mechanical tactile estimation result and the data-driven tactile estimation result are used as multi-source observation information. State estimation and filtering methods are introduced for fusion calculation. Dynamic modeling and compensation are performed on the dynamic effects caused by the flexible covering layer to obtain the fused tactile perception result. Step S5: Output the fused tactile perception result, which includes the magnitude and direction of the three-dimensional contact force and the coordinates of the contact position.

2. The multi-dimensional tactile sensing method for robot fingertips according to claim 1, characterized in that, The state estimation and filtering method in step S4 is specifically the Kalman filtering method, which recursively estimates the tactile state by constructing a state space model.

3. The robot fingertip multidimensional tactile sensing method according to claim 1, characterized in that, The machine learning model in step S3 is a multilayer perceptron model, which is used to realize the nonlinear end-to-end mapping from multi-channel raw tactile data to three-dimensional contact force and contact position.

4. The robot fingertip multidimensional tactile sensing method according to claim 1, characterized in that, The principle of force and torque conservation in step S2 specifically includes: based on the principle of force balance and the mapping method of the center of pressure force application point, the multi-channel strain signal is analyzed and calculated to obtain the three-dimensional force components and contact position.

5. The robot fingertip multidimensional tactile sensing method according to claim 1, characterized in that, The rigid-flexible hybrid structure robot fingertip tactile sensing device includes: a flexible covering layer located on the outermost layer for direct contact with objects, a rigid support structure fixedly connected to the flexible covering layer, and a plurality of multi-axis strain sensing units arranged on the rigid support structure.

6. The robot fingertip multidimensional tactile sensing method according to claim 1, characterized in that, The three-dimensional contact force magnitude and direction, as well as the contact position coordinates output in step S5, are directly used for robot gripping force adjustment or posture control, forming a closed-loop control driven by tactile feedback.

7. The robot fingertip multidimensional tactile sensing method according to claim 1, characterized in that, The dynamic effects in step S4 include low-pass hysteresis caused by the flexible overlay, slow-varying bias, and random noise; the dynamic modeling and compensation are used to suppress the above effects and improve the temporal consistency and long-term stability of the tactile signal.

8. The robot fingertip multidimensional tactile sensing method according to claim 1, characterized in that, In step S4, a state-space modeling approach is adopted, which treats the external contact force and contact position as system states that change over time, and recursively estimates the system states in a continuous time series.

9. The robot fingertip multidimensional tactile sensing method according to claim 1, characterized in that, The machine learning model used in step S3 is trained in advance using a dataset and is used to compensate for deviations in the results of the physical perception calculation steps at the edge of the force-bearing area and within a specific force amplitude range.

10. The robot fingertip multidimensional tactile sensing method according to claim 1, characterized in that, The three-dimensional components of the external contact force calculated in step S2 include the normal force and the tangential force; the contact position coordinates are the two-dimensional coordinates of the contact point on the plane where the rigid support structure is located.

Images