Tactile sensor and method of operating the tactile sensor
By using a classification scheme generated by multiple stress sensors and machine learning algorithms in the haptic sensor, the problem of insufficient speed and versatility in the prior art is solved, and efficient detection and classification of static and dynamic haptic forces is achieved.
Patent Information
- Application Number
- CN202080045505.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2019-06-24
- Filing Date
- 2020-06-22
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2040-06-22
AI Technical Summary
Existing haptic sensors have shortcomings in speed and versatility, making it difficult to effectively detect static and dynamic haptic forces, especially in high-speed changing load modes.
The Wheatstone bridge structure consisting of multiple stress sensors is combined with the decision tree integrated classification scheme generated by the machine learning algorithm. The static and dynamic tactile forces on the surface are detected through the stress sensor and classified into predefined load mode categories.
It realizes efficient detection and classification of static and dynamic haptic forces, and can accurately distinguish different load modes in the frequency range of 1Hz to 1000Hz, improving the speed and versatility of the haptic sensor.
Smart Images

Figure CN113994181B_ABST
Abstract
Description
[0001] A tactile sensor is provided. In addition, a method for operating the tactile sensor is also provided.
[0002] One object to be achieved is, in particular, to specify a tactile sensor with improved speed and versatility. Another object to be achieved is, in particular, to specify a method for operating a tactile sensor with improved speed and versatility.
[0003] A tactile sensor is a transducer by means of which a load pattern on a detection surface of the tactile sensor can be detected. The load pattern may be caused by an object sliding along the detection surface, an object statically pressing against the detection surface, or an object bouncing off the detection surface, or a mixture of the above processes.
[0004] The tactile sensor includes a plurality of stress sensors. The stress sensors may be arranged on a common substrate. In particular, the stress sensors may be manufactured using a common manufacturing method, in particular using a C-MOS process. In particular, the stress sensors are integrated into a general-purpose chip. For example, the stress sensors may have been manufactured using a common manufacturing process. The stress sensors may be composed of p-MOS and / or n-MOS transistors, respectively. For example, each stress sensor includes four transistors connected as a Wheatstone bridge.
[0005] The tactile sensor comprises at least one contact body. The contact body may be a whole body in which the stress sensor is embedded. In particular, the surface of the stress sensor that is sensitive to mechanical stress is completely covered by the contact body. For example, the contact body provides a mechanical connection between the stress sensor and the object causing the load pattern.
[0006] The stress sensors are used to detect static tactile forces and dynamic tactile forces applied to the detection surface of the contact body. In particular, the contact body transfers mechanical stresses from its detection surface to the stress sensors. The stress sensors can be mechanically connected to a common contact body. In particular, the detection surface of the contact body is mechanically connected to all stress sensors of the contact body. Both dynamic and static tactile forces can act perpendicularly or obliquely on the detection surface.
[0007] A static tactile force is a constant force acting on the detection surface. In this case, the constant force does not change in magnitude or direction over a first time span, wherein the first time span is at least one hundred times, preferably at least one thousand times, the maximum sampling rate of the tactile sensor. For example, the first time span is at least 0.5 seconds.
[0008] The dynamic tactile force is a changing force acting on the detection surface. In this case, the changing force changes in magnitude and / or direction over a second time span, wherein the second time span is at least twice the sampling rate of the tactile sensor. For example, the second time span is at most five seconds, preferably at most 0.5 seconds.
[0009] For example, the dynamic tactile force varies with a frequency of 1 Hz to 1000 Hz. Advantageously, the detection of tactile forces of 1 Hz to 1000 Hz enables the detection of surface characteristics of an object sliding along the detection surface.
[0010] In particular, the dynamic tactile force varies with a frequency of 40 Hz to 400 Hz. Advantageously, the detection of the dynamic tactile force with a frequency of 40 Hz to 400 Hz enables the detection of an object sliding along the detection surface.
[0011] The chip includes a memory having a classification scheme, wherein the classification scheme assigns a set of output values of the plurality of stress sensors to a predefined load pattern.
[0012] The classification scheme is generated by a machine learning algorithm. The output of the machine learning algorithm is a classification scheme, which can be a decision tree ensemble algorithm. In particular, the classification scheme is a random forest algorithm.
[0013] Furthermore, a method for generating a classification scheme is specified. By means of this method, in particular, a classification scheme for a tactile sensor described here can be generated. Therefore, all features disclosed for the tactile sensor are also disclosed for this method, and vice versa.
[0014] For example, a classification scheme, also called training of a decision tree ensemble, is obtained by performing the following steps a) to c):
[0015] a) Predefined load pattern categories. By predefining the categories of load patterns, possible load patterns can be predetermined by means of the tactile sensor. Different load pattern categories can be distinguished as an object sliding along the detection surface, bouncing off the detection surface or pressing against the detection surface. Furthermore, different load pattern categories can be distinguished by parameters such as the material of the object, the surface characteristics of the object, the speed at which the object slides along the detection surface, the pressure of the object on the detection surface.
[0016] b) using a reference tactile sensor to perform multiple representative measurements for each predefined load pattern category. For example, the load pattern for each predefined load pattern category is measured at least 100 times, preferably at least 300 times. In this measurement, each stress sensor provides an output value in response to the applied load pattern. The output value of the stress sensor can be a voltage value or a current value, wherein the voltage or current depends on the amount of stress sensed by each sensor, respectively.
[0017] c) defining a decision tree ensemble by assigning a set of output values to each predefined load pattern category, respectively. In this context, a set of output values is the totality of all output values of the plurality of stress sensors in response to the load pattern. After performing a plurality of representative measurements, a plurality of sets of output values are assigned to each load pattern category.
[0018] A classification scheme obtained by a plurality of measurements with the representative tactile sensor is loaded into the memory of the tactile sensor. The classification scheme is capable of assigning a set of output values caused by a load pattern to one of the predefined load pattern categories. Thus, the predefined load pattern categories are detected with the tactile sensor.
[0019] In particular, the classification schemes for the load pattern categories generated by static haptic forces and dynamic haptic forces are generated by the same method and the same settings.
[0020] Furthermore, a method for operating a tactile sensor is specified. By means of this method, in particular the tactile sensor described here can be produced. Therefore, all features disclosed for the tactile sensor are also disclosed for the method, and vice versa.
[0021] A method for operating a tactile sensor comprising a contact body having a detection surface, a plurality of stress sensors and a chip comprises the following method steps:
[0022] - applying a load pattern on the detection surface of the contact body;
[0023] -Converting the load pattern into a set of output values with the help of stress sensors;
[0024] - Assignment of output values to predefined load pattern categories with the help of chips and classification schemes.
[0025] For example, the detection surface is one or more surfaces of the contact body. In particular, the detection surface is freely accessible. Thus, the object may come into direct contact with the contact body.
[0026] The load pattern may be caused by an object in direct contact with the detection surface. The load pattern may be a dynamic tactile force or a static tactile force, which is transmitted to the plurality of stress sensors through the contact body. The stress sensors return output values respectively in response to the load pattern.
[0027] The total output values returned in response to the load pattern are a set of output values. The set of output values can be assigned to one of the predefined load pattern categories with the aid of a decision tree ensemble. The tactile sensor can return a predefined load pattern category to which the output value is assigned when detected.
[0028] The load pattern is generated by static tactile forces and / or dynamic tactile forces on the detection surface. The static tactile forces and the dynamic tactile forces on the detection surface are classified by means of the same classification scheme. In particular, the load pattern consists of a mixture of dynamic and static tactile forces. Both dynamic and static tactile forces are classified by means of the same stress sensor and the same classification scheme. In particular, the load pattern is detected and classified without performing a spectral analysis of the output values.
[0029] Further advantages and advantageous embodiments as well as further developments of the tactile sensor and of the method for operating a tactile sensor will become apparent from the following description of the exemplary embodiments in conjunction with the drawings.
[0030] Figure 1 Exemplary embodiments of tactile sensors and chips are shown;
[0031] Figure 2 An exemplary embodiment of a method for manufacturing a tactile sensor is shown;
[0032] Figure 3A and 3B An exemplary embodiment and a schematic embodiment of a chip used in a tactile sensor are shown;
[0033] Figure 4A and 4B An exemplary embodiment of a stress sensor used in a tactile sensor is shown;
[0034] Figure 5 , 6 and 9 show an exemplary embodiment of a tactile sensor having a load pattern applied thereto;
[0035] Figure 7 and 8 a scatter plot showing output values of an exemplary stress sensor;
[0036] Fig.10 showing a scatter plot of exemplary strain sensor output values in response to a load pattern having static and dynamic tactile forces;
[0037] Fig.11 shows output values of an exemplary stress sensor and classifications resulting from those output values;
[0038] Fig.12 and 13 shows a 3D plot of an exemplary set of output values relative to the chip surface;
[0039] Fig.14 An exemplary measurement setup for generating a classification scheme is shown;
[0040] Fig.15shows exemplary output values of a single stress sensor under different load modes;
[0041] Fig.16A , 16B 16C show an exemplary embodiment of a set of decision trees used in a tactile sensor.
[0042] The same, similar or equivalent elements are provided with the same reference numerals in the drawings. The drawings and the ratios between the elements shown in the drawings should not be considered to be drawn to scale unless there is a clear unit description. On the contrary, the size of the individual elements can be exaggerated for better presentation and / or better understanding.
[0043] Figure 1 An exemplary embodiment of a touch sensor 1 and an exemplary embodiment of a chip 10 used in the touch sensor 1 are shown. The touch sensor 1 comprises a contact body 200 in which the chip 10 is embedded. A mount 201 can electrically connect the chip 10 by means of a flexible cable. In this exemplary embodiment, the mount 201 is embedded in the contact body 200.
[0044] The chip 10 comprises a plurality of stress sensors 100.1, 100.2, 100.3, 100.4. The chip 10 comprises 32 stress sensors and all necessary readout circuits. The chip is bonded to a flexible flat cable for power supply and serial communication. Advantageously, the chip is compact and easy to handle.
[0045] To form the tactile sensor 1, the chip 10 is embedded in a contact body 200 made of silicone (PDMS). The contact body 200 has a fingertip shape. Advantageously, the chip does not require any additional electronic components and wiring.
[0046] The load pattern on the contact body 200, especially the load pattern on the detection surface 200a, deforms the contact body and induces stress distribution in the chip 10. 32 stress sensors 100 are distributed over the chip area and can measure the stress distribution in the chip. The stress distribution is specific to the direction and intensity of the load pattern.
[0047] A possible load pattern can be a dynamic tactile force caused by an object 300 sliding along the detection surface 200a. The tactile sensor 1 is capable of measuring vibrations up to 400 Hz, which is essential for detecting such load patterns. Therefore, in order to ensure sufficient bandwidth, the tactile sensor 1 has a sampling rate of at least 960 Hz. Since the stress distribution of the entire chip is measured, all 32 stress sensors are read, which reduces the sampling rate to 30 Hz. The stress sensors are read serially, so the vibrations are still sampled at 960 Hz. Therefore, the sampling rate of each stress sensor 100 corresponds to the sampling rate of each tactile sensor 1 divided by the number of stress sensors 100 per chip 10.
[0048] The vibrations are recorded at multiple points on the surface of the chip 10. The spatial distribution of the stress sensors 100 on the chip 10 has a negligible effect on the measurement, as long as the measured wavelength of the vibration is much larger than the distance of the stress sensors 100 on the chip 10. For example, the maximum distance of the stress sensors 100 on the chip 10 is at least 1 / 100, preferably at least 1 / 1000, and more preferably at least 1 / 100000 of the wavelength of the vibration to be measured.
[0049] Figure 2 An exemplary method for manufacturing a tactile sensor 1 is shown. In a first method step, a mold 203 is filled with a first silicone layer. The bottom side of the silicone layer facing the mold 203 becomes the detection surface 200a of the tactile sensor 1 .
[0050] In a second step, the chip 10 is laid flat on top of the first silicone layer, on the side facing away from the mold 203. The silicone has sufficient viscosity to hold the chip in place.
[0051] In a third method step, the mounting element 201 is fixed to the mold 203 .
[0052] In a fourth method step, the contact body 200 is completed by filling the mold with a silicone material. In this method step, the chip 10 is completely embedded in the contact body 200. The mount 201 is partially embedded in the contact body 200.
[0053] In a fifth method step, the contact body 200 is cured at a pressure of 2 bar and 60° C. The contact body 200 is made of PDMS (Dowsil 3140). The PDMS material advantageously provides a tight mechanical coupling between the stress sensor 100 and the contact body 200 .
[0054] In a sixth method step, the mold 203 is removed from the contact body 200. After removal of the mold 203, the detection surface 200a is freely accessible.
[0055] Figure 3Aand 3B A top view and schematic diagram of an exemplary chip 10 used in the tactile sensor 1 are shown. The chip is manufactured in a 0.35 μm process. The chip 10 includes 24 transistor-based stress sensors 100, which are strategically distributed over the chip area to measure the difference between the in-plane shear stress and the in-plane normal stress. The stress sensor signals are processed by a variable gain differential amplifier and digitized by a 10-bit SAR ADC. The chip 10 has a resolution as low as 11 kPa at the highest gain.
[0056] 24 stress sensors 100 are connected serially to the readout unit through a 5-bit multiplexer (MUX). The processed stress sensors 100 can be biased in four directions through an additional multiplexer, which makes offset compensation operation possible due to the Wheatstone bridge architecture of the stress sensor. A variable gain differential difference amplifier (DDA) together with a 10-bit successive approximation (SAR) analog-to-digital converter (ADC) performs the stress sensor readout.
[0057] like Figure 3A As shown, the area of the chip 10 is 2×2.5 mm^2.
[0058] like Figure 3B The illustrated stress sensor 100 reader consists of a differential difference amplifier (DDA) with variable gain, followed by a SAR ADC and a gain controller (GC). Based on the digitized output value of the stress sensor, the gain controller determines the maximum ideal gain value relative to the linear DDA output range based on a binary tree search.
[0059] The architecture of the DDA provides a differential input pair for the stress sensor and a separate differential input pair for the feedback. Thus, a high-ohmic interface for the stress sensor connection is provided, and impedance changes in the feedback path during gain changes do not affect the load impedance of the stress sensor during readout.
[0060] Figure 4A and 4B A top view of a stress sensor 100 is shown. Each stress sensor forms a square active element and can be described by four transistors in a Wheatstone bridge arrangement. The orientation is indicated by a coordinate system (x, y) parallel to the edge of the chip 10. The doping polarity has an equally strong influence on the piezoresistance coefficient of the MOS inversion layer. For example, the π in p-type silicon 44 is ten times that of n-type silicon, while n-type silicon has a 12is nearly 50 times greater than in p-type silicon. Therefore, the in-plane shear stress is measured with the NMOS type stress sensor 100 rotated 45° with respect to (x,y). The PMOS type stress sensor is oriented parallel to the coordinate system and is therefore sensitive to differences in in-plane normal stress. All four transistors within each stress sensor have a W / L ratio of 5μm / 10μm. These dimensions represent a good compromise between the four main sensor characteristics.
[0061] Each NMOS stress sensor is activated by connecting the common gate of its transistor to the drain voltage. Each PMOS stress sensor is activated by connecting the common gate of its transistor to the source voltage. This reduces the power consumption of the entire system because only the processed sensors are temporarily activated.
[0062] Figure 5 and Figure 6 Schematic side and top views of a tactile sensor 1 according to an exemplary embodiment are shown. The tactile sensor 1 includes a detection surface 200a in direct contact with an object 300. A vertical static tactile force FS0, a forward static tactile force FS1, a backward static tactile force FS2, and lateral static tactile forces FS4, FS3 between the object 300 and the tactile sensor 1 are detectable load patterns.
[0063] Figure 7 and Figure 8 The measurement results of four stress sensing load patterns with static tactile forces in different directions are shown as a scatter plot. For this measurement, category data is used to assign a set of measured output values to predefined load pattern categories. Each point in the scatter plot represents a scatter sample and is color-coded according to the load category assigned to it.
[0064] The reference data marked as "no contact" were collected without any mechanical contact between the object 300 and the contact body 200. The data with pure normal static tactile force is marked as FS0. In addition to the normal static tactile force, a tangential static force is applied by gradually moving the object 300 under the tactile sensor 1. Figure 5 and Figure 6 The data are labeled FS1, FS2, FS3, and FS4 according to the corresponding directions shown in FIG.
[0065] Classification schemes, in particular, decision tree ensembles, are trained to detect contacts and classify loading patterns caused by shear forces in different directions.
[0066] Depending on the strain sensor, certain load pattern classifications result in similar output values for the strain sensor. However, the overall output value, the set of output values, is distinguishable for different load patterns. The classification scheme takes into account the entire set of output values so that the applied load pattern is reliably assigned to the predefined load pattern class. The trained random forest algorithm can assign the load pattern to the correct load pattern class with 99.8% accuracy.
[0067] In addition, if Fig. 9 As shown, the load mode category can be defined as the tactile sensor 1 sliding along the object 300 with its detection surface 200a. The detection surface 200a can have a surface structure with a plurality of parallel ridges. When the object 300 slides along the detection surface 200a, the ridges cause vibrations of the contact body 200. The detection of vibrations makes it possible to distinguish between static tactile forces and dynamic tactile forces.
[0068] By measuring the vibrations with a sampling rate of 960 Hz by means of a single strain sensor of chip 10 , it can be ruled out that the entire vibration is damped by the contact body. Fig.15 The sensor output in three different states A, B and C is shown. In state A, the tactile sensor 1 does not move relative to the object 300. However, in state A, there is only a static tactile force perpendicular to the detection surface. In state B, the tangential force increases until the object 300 slides along the detection surface 200a. In state C, the object 300 slides along the detection surface 200a at a speed of 10 mm / s and causes the output value of the stress sensor 100 to oscillate. Therefore, the dynamic tactile force can be easily detected by the spectrum analysis of the output value of a single sensor.
[0069] By analyzing the outputs of multiple strain sensors 100 , and in particular all 32 strain sensors, the load patterns of static and dynamic tactile forces may be detected without the need for spectrum analysis. Fig.10 The complete classification scheme is shown as a scatter plot of the two most important stress sensors 100. Each point in the scatter plot represents a sensor sample and is color-coded according to the predefined load category assigned to the corresponding output value. The random forest classification algorithm was trained using at least 50 trees and unconstrained tree growing. The overall accuracy of 5-fold cross validation was 99.6%. At least 98% truth rate was achieved for each load pattern type.
[0070] There is no sensor size where the output value sets in response to static and dynamic tactile forces do not overlap. Therefore, reliable classification of load patterns with the aid of one or two strain sensors is not an easy task. Dynamic tactile forces may only be classified by the relationship of the output values of multiple strain sensors.
[0071] Fig.12The 3D graph shown shows the output values of the stress sensor 100 on the surface of the chip 10 when a static tactile force is used as the load mode. Fig.13 The 3D plot shown shows the output values of the stress sensor 100 on the surface of the chip 10 when a dynamic tactile force is used as the load pattern. For this visualization, the data is high-pass filtered to suppress any offset patterns from the directional static force. Deviations from the plane in the static plot are due to sensor noise. Therefore, static tactile forces lead to well-defined stress patterns that can be distinguished from distortion patterns in sliding samples. Through training, the classification algorithm can learn to distinguish these patterns into two different load pattern categories, making it possible to detect dynamic and static tactile forces with the help of the same algorithm.
[0072] Fig.11 A time domain measurement with multiple stress sensors 100 is shown to visually verify the classification. The output values of the four stress sensors are shown during the following load patterns: pure static normal tactile force FS0, static tangential tactile force FS1 or FS2, forward dynamic tactile force FD1. The sequence in the measurement shown is: pure normal force at the beginning of the measurement for 2.5 seconds, static forward force increases until 4.7 seconds of the measurement, the object slides along the detection surface 200a until 6.8 seconds, the sliding movement stops until 9.5 seconds, and the sliding movement of the object along the detection surface 200a starts again. The figure also shows the classification results for each sensor sample.
[0073] Whenever a sliding movement causes a vibration of the contact body 200, the vibration results in an oscillating signal of the stress sensor. The classification scheme enables the detection and correct classification of the load pattern of the sliding movement substantially correctly and reacting very quickly to the sliding movement.
[0074] Fig.14 A setup for generating a classification scheme is shown. The setup comprises a classification mount 202 to which a tactile sensor 1 is attached. With the aid of the classification mount 202, the tactile sensor is pushed against an object 300 with a definable normal tactile force FS0. The object 300 is a 3D printed plate with different surface patterns. The surface pattern can be smooth, ridged, circular or triangular. The surface pattern allows the classification scheme to be generalized from the surface. Otherwise, the classification algorithm will be limited to classifying the load patterns of objects with only one specific surface structure. The object 300 is mounted on an XY cross stage. The object 300 can be moved parallel to the detection surface 200a by means of a stepper motor. These steps are fine enough to gradually increase the tangential tactile force without sliding the tactile sensor on the object. The stepper motor can accelerate and move the workbench at a speed of up to 10 mm / s to produce sliding. The tangential force applied to the detection surface is recorded by the verification sensor 5.
[0075] When generating a classification scheme, a predefined load pattern is applied to the tactile sensor 1 and the corresponding output values are assigned to the predefined load pattern categories. This measurement is repeated multiple times to generate a reliable decision tree ensemble.
[0076] Fig.16A An example of a single decision tree after training is shown. In this context, training is the measurement of predefined load patterns and assigning corresponding output values to load pattern classes. The feature f of a node i and threshold t i , the leaf category label C i , the entire tree structure will change during the training process. In the context of tactile sensor 1, stress sensors 100.1, 100.2, 100.3, 100.4 correspond to features f1, f2, f3, f4, and threshold t i Output value of corresponding stress sensor 100, category C i Corresponds to predefined load pattern categories.
[0077] Fig. 16B A static implementation in the parallel architecture section is shown. In order to be able to continuously train a decision tree, the tree structure must remain reconfigurable.
[0078] like Fig. 16C As shown, the tree structure can be implemented by reconfigurable counter parts commonly used in FPGAs. The assignment of features to nodes is achieved through a switching and interconnection network that connects the feature inputs to the appropriate node comparators. The Boolean functions that encode the tree structure and calculate the leaf outputs are replaced by lookup tables (LUTs) used in FPGA logic blocks. Only the comparators remain fixed, and their thresholds must be stored in registers. This enables the architecture to support updates of the decision tree.
[0079] The full implementation of a decision tree ensemble includes switching and interconnect blocks, which span the ensemble and allow sharing of comparators between trees. In particular, the trees of a decision tree ensemble can be of different sizes.
[0080] The invention is not limited by the description based on these embodiments. On the contrary, the invention includes any novel feature and any combination of features, in particular including any combination of features in the claims, even if this feature or combination itself is not explicitly stated in the patent claims or exemplary embodiments.
[0081] Reference numerals list
[0082] 1. Tactile Sensor
[0083] 10 Chip
[0084] 100 Strain Sensors
[0085] 100.1 First stress sensor
[0086] 100.2 Second stress sensor
[0087] 100.3 Third stress sensor
[0088] 100.4 Fourth stress sensor
[0089] 111 Classification Scheme
[0090] 101 Output value set
[0091] 110 Memory
[0092] 200 Contact
[0093] 201 Mounting parts
[0094] 202 Assorted mounting parts
[0095] 203 Mould
[0096] 200a Test surface
[0097] 300 objects
[0098] 5 Verify the sensor
[0099] FS Static tactile force
[0100] FS0 Upward static tactile force
[0101] FS1 Forward Static Haptic Force
[0102] FS2 Backward Static Haptic Force
[0103] FS3 Right Static Haptic Force
[0104] FS4 Left Static Haptic Force
[0105] FD Dynamic Haptic Force
[0106] FD1 Forward Dynamic Haptic Force
[0107] FD2 Backward Dynamic Haptic Force
Claims
1. A tactile sensor (1), comprising: - a chip (10) comprising a plurality of stress sensors (100) and a memory (110) having a classification scheme (111), and - at least one contact body (200), in, The stress sensor (100) is integrated into the chip (10) and distributed over an area of the chip (10). The stress sensor (100) is configured to detect a load pattern applied to a detection surface (200a) of a contact body (200). wherein the classification scheme (111) assigns a set of output values (101) of the plurality of stress sensors (100) to a predefined load pattern, The classification scheme (111) is a decision tree ensemble algorithm and is generated by means of a machine learning algorithm, and The classification scheme (111) is generated by the following steps: a) Categories of predefined load patterns; b) performing a number of representative measurements for each predefined load pattern category using a reference tactile sensor; and c) Defining a decision tree ensemble by assigning a set of output values to each predefined load pattern category respectively.
2. The tactile sensor (1) according to claim 1, wherein: The load modes include static tactile force (FS) and / or dynamic tactile force (FD), and The dynamic tactile force (FD) varies at a frequency of 1 Hz to 1000 Hz.
3. A method for operating a tactile sensor (1), the tactile sensor comprising a contact body (200) having a detection surface (200a) and a chip (10) comprising a plurality of stress sensors (100) and a memory (110) having a classification scheme (111), wherein: The stress sensor (100) is integrated into the chip (10) and distributed over an area of the chip (10). in, - applying a load pattern on the detection surface (200a) of the contact body (200), - the stress sensor (100) converts the load pattern into a set of output values (101) by means of the stress sensor (100), and - the chip (10) assigns the output value (101) to a predefined load pattern class with the aid of the chip (10) and the classification scheme (111), The classification scheme (111) is a decision tree ensemble algorithm and is generated by means of a machine learning algorithm, and The classification scheme (111) is generated by the following steps: a) Categories of predefined load patterns; b) performing a number of representative measurements for each predefined load pattern category using a reference tactile sensor; and c) Defining a decision tree ensemble by assigning a set of output values to each predefined load pattern category respectively.
4. The method according to the preceding claim, wherein: The load patterns generated by static and dynamic tactile forces on the detection surface are classified by the same classification scheme.
5. The method according to claim 3 or 4, wherein: The dynamic tactile force is classified without performing a spectral analysis on the output value (101).
6. The method according to claim 3 or 4, wherein: The generating of the classification scheme also includes generating a classification scheme for classifying load patterns generated by the static haptic forces and the dynamic haptic forces.
