Search

CN-121437772-B - Model construction method of VEM-4D dynamic skin structure and dynamic expression

CN121437772BCN 121437772 BCN121437772 BCN 121437772BCN-121437772-B

Abstract

The invention relates to the field of artificial intelligence, and discloses a structure of a VEM-4D dynamic skin and a construction method of a dynamic expression model, and provides a 4D dynamic skin structure, under the control of a 4D expression model and a driving module, dynamic expression of the skin is simulated, wherein an expression function vector is generated in the 4D expression model by an expression instruction, exciting current is provided by the driving module to drive a parallel electromagnet matrix arranged in an elastic soft magnetic silica gel composite material, so that two-dimensional surface expansion deformation of the skin is realized, concave-convex electromagnets are driven, three-dimensional concave-convex deformation of the skin is realized, time-saving current beats are adopted to realize beat peristaltic motion of the skin, and a mapping set and an expression library corresponding to the expression function vector and the dynamic expression are obtained by adopting supervised learning and reinforcement learning.

Inventors

  • DING XIANGEN
  • DING YUANTONG

Assignees

  • 港湾之星健康生物(深圳)有限公司

Dates

Publication Date
20260512
Application Date
20251229

Claims (10)

  1. A vem-4D dynamic skin structure comprising: the VEM-4D dynamic skin includes a 4D dynamic skin and a drive module, wherein, The 4D dynamic skin comprises more than 2 electromagnets, the electromagnets comprise parallel electromagnets formed by soft magnetic materials and exciting coils, the parallel electromagnets are arranged side by side according to an interval A, the front surface of the 4D dynamic skin comprises a surface skin adhered to the front magnetic poles of the parallel electromagnets, and the surface skin is formed by adopting an elastic telescopic membranous shape; the driving module generates attraction force or repulsion force between magnetic poles at two ends of the electromagnet by applying exciting current to the exciting coil, and drives the epidermis to generate telescopic deformation; The VEM-4D dynamic skin also comprises a 4D expression model, and the 4D expression model provides static and dynamic expression instructions for the driving module to drive the epidermis to stretch and deform.
  2. 2. The structure of claim 1, further comprising the structure of: The 4D dynamic skin comprises a transverse spacer with a through hole buckle, the through hole buckle clamps the back magnetic poles of the parallel electromagnets, and a space B is reserved between the through hole buckle and the back magnetic poles; the transverse diaphragm is made of elastic material sheet, and its vertical direction can produce vertical elastic deformation when it is stressed, and/or, The electromagnet further comprises an unsmooth electromagnet which is positioned outside the back magnetic pole and the transverse spacer and is provided with a space C, the unsmooth electromagnet comprises soft magnetic materials and exciting coils and is arranged on a backboard which is arranged on the 4D dynamic skin, the exciting coils of the unsmooth electromagnet are driven by a driving module through expression instructions, so that the magnetic pole of the unsmooth electromagnet generates attraction force or repulsive force to more than one back magnetic pole, the epidermis is driven to generate unsmooth deformation, and/or, The arrangement of the electromagnets and the shape of the 4D dynamic skin are produced according to practical application, and a transverse spacer support column and a surface positioning column are further arranged between the transverse spacer and the back plate.
  3. 3. The structure of claim 2, further comprising the structure of: The 4D dynamic skin further includes a subcutaneous tissue of an elastic substance filled around the parallel electromagnet, and a skin is adhered to a surface of the subcutaneous tissue on the front side of the 4D dynamic skin, in which case the skin is deformed to be stretched and deformed to be concave and convex by the subcutaneous tissue due to suction or repulsive force of the front side magnetic pole and the action of the concave-convex electromagnet, and/or, The subcutaneous tissue may also include a powder of homogeneously mixed soft magnetic material therein and/or, The subcutaneous tissue also comprises cavities which are communicated with each other, wherein the cavities are filled with heat dissipation fluid, and a temperature sensor is arranged so as to control the temperature of the epidermis within a set range; The subcutaneous tissue wraps and seals the skin to form a complete 4D dynamic skin, and an inlet and an outlet for heat dissipation fluid are arranged.
  4. 4. A structure according to claim 3, further comprising the structure: The driving module comprises an input end, a coder-decoder, a current source and an output end, wherein the output end is connected with two ends of an excitation coil of each electromagnet, the current source provides excitation current, the coder-decoder provides decoding according to the codes of the electromagnets, the input end is connected with a 4D expression model and a driving power supply, the 4D expression model drives the electromagnets to generate telescopic expression and concave-convex expression on the epidermis through the driving module, and/or, The current source comprises a constant current source, a pulse current source, a modulated pulse power source, and/or, The driving module is also connected with more than one temperature sensor, the temperature sensors are arranged at the appointed positions of the subcutaneous tissues, and temperature signals are transmitted back to the 4D expression model so as to monitor the temperature of the subcutaneous tissues; the driving module also comprises a connecting terminal for driving a circulating pump of the heat dissipation fluid, and the temperature control and the circulation of the heat dissipation fluid are completed under the control of the 4D expression model, and/or, The driving module further comprises a filter circuit for reducing or eliminating transient oscillation of a magnetic field in the arrangement of the electromagnets generated by the high-frequency pulse of the 4D expression model; the drive module further comprises a degaussing current for providing degaussing to the electromagnet under control of the 4D expression model and/or, The driving module further comprises a wireless communication sub-module which is in external communication with the 4D dynamic skin under the control of the 4D expression model.
  5. The VEM-4D dynamic expression model construction method is characterized by comprising the following steps: The VEM-4D dynamic expression model comprises a 4D expression model, a 4D dynamic skin and a driving module, wherein an expression instruction is loaded at the input end of the 4D expression model, an expression function vector generated at the output end provides exciting current for an exciting coil of an electromagnet in the 4D dynamic skin through the driving module and controls the exciting current, and suction force or repulsive force between the magnetic poles of more than 2 electromagnets drives a epidermis positioned on the front surface of the 4D dynamic skin to generate beat peristaltic motion so as to simulate the dynamic expression; S2000, the expression instruction includes a telescopic instruction, acts on parallel electromagnets arranged side by side included in the electromagnets, simulates a telescopic expression included in a dynamic expression in the epidermis, and/or, And S3000, the expression instruction comprises a concave-convex instruction, the concave-convex instruction acts on a concave-convex electromagnet arranged on the back surface of the 4D dynamic skin and is included in the electromagnet, a magnetic pole arranged on the back surface of the 4D dynamic skin and is driven by the parallel electromagnet through an elastic transverse diaphragm, and the concave-convex expression included in the dynamic expression is simulated in the epidermis.
  6. 6. The method according to claim 5, characterized in that S1000 comprises in particular the following methods: S1100, the expression instruction comprises an expression function FE (x, y, z, t, i, p, v and tau), wherein x, y, z and t in expression function vectors (x, y, z, t, i, p, v and tau) are respectively coordinates of a 3-dimensional space of an electromagnet and coordinates of a beat value of a 4 th dimension, i, p, v and tau are respectively a current value, a current polarity, a type of the electromagnet and a step length time of the electromagnet, the step length time is larger than the beat value, the time length of transition from a previous expression to a next expression in a dynamic expression is longer, and a 4D expression model dynamically drives a epidermis to simulate a sequence of the dynamic expression according to front and back changes of each group of vector sequences in the expression instruction; S1200, wherein the current source comprises a constant current source, a pulse current source and a modulated pulse power source, the 4D expression model comprises serial programming and parallel programming of more than 2 exciting coils through a driving module, and the method specifically comprises the following steps of S1210 and S1220: the series programming is operated in a current source mode, and when having the same current value, beat value and current polarity, the exciting coils are operated in series through the driving module, and/or, S1220 that the parallel programming is operated in a constant voltage source mode, when having the same voltage value, beat value and current polarity, the exciting coils are operated in parallel by the driving module, and/or, The expression instruction also comprises an expanded system temperature to drive a temperature control system comprising the 4D dynamic skin, wherein the temperature control system comprises a temperature sensor, a heat dissipation fluid, a heat dissipation heater and a circulating pump, the 4D expression model performs temperature control through a driving module according to the system temperature in the expression instruction, and/or, S1400, the expression instruction comprises a VEM expression vector instruction supporting voice emotion multi-mode, wherein the VEM expression vector instruction comprises an expression library from artificial intelligence supervised learning and reinforcement learning.
  7. 7. The method according to claim 6, wherein S1000 specifically further comprises the following method: s1500, calculating absolute values F 1 of attractive force or repulsive force between the magnetic poles of adjacent parallel electromagnets with the spacing sp between the edges of 2 magnetic poles and the radius gamma of the magnetic dipoles by adopting a magnetic dipole model, calculating a magnetic moment saturation current I 1 , selecting the stretching elastic force F 3 of the elastic material of the epidermis, and F 3 ≤F 1 , determining that the exciting current of an exciting coil is more than or equal to I 1 , and ensuring that the attraction or repulsion of the magnetic poles causes the epidermis to produce beat peristaltic motion; S1600, calculating more than 2 parallel electromagnets according to a magnetic dipole model, wherein when the polarity of exciting current in each exciting coil changes, attraction or repulsion of magnetic poles enables the epidermis to generate a process and a result of beat peristaltic motion, and the result of beat peristaltic motion enables the epidermis to generate dot-shaped expansion, linear expansion and block-shaped expansion so as to simulate dynamic expression, and/or, S1700, according to a magnetic dipole model, adopting a concave-convex electromagnet to calculate the current I 2 of the concave-convex electromagnet reaching magnetic moment saturation, wherein the absolute value F 2 of suction force or repulsive force generated by the magnetic poles of the concave-convex electromagnet on the back of the magnetic pole pair of the parallel electromagnet, the tension of the transverse diaphragm is F 4 , F 4 ≤F 2 is selected, and F 2 ≥F 1 causes the transverse diaphragm to deform, the parallel electromagnet is pushed and pulled to cause the epidermis to generate projections and depressions, the epidermis to generate punctiform projections and depressions, linear projections and depressions and blocky projections and depressions so as to simulate dynamic expression, and/or, And S1800, the skin comprises elastic materials and/or subcutaneous tissues uniformly mixed with soft magnetic material powder in the 4D dynamic skin, wherein the expansion elasticity of the subcutaneous tissues is smaller than or equal to the difference value of the absolute value of the suction force or the repulsive force between the magnetic poles of the parallel electromagnets and the expansion elasticity of the skin elastic materials, so that the suction force or the repulsive force between the magnetic poles is ensured to be enough to resist the expansion force of the subcutaneous tissues and the skin, and the skin generates beat peristaltic motion.
  8. 8. The method according to claim 7, further comprising a supervised learning method of S4000, comprising in particular: s4100, driving the parallel electromagnets to generate punctiform extension, linear extension and block extension on the epidermis according to the expression instructions, driving the concave-convex electromagnets to generate punctiform protrusions and depressions, linear protrusions and depressions and block protrusions and depressions on the epidermis, manually judging and establishing a group of mapping of expression function vectors and a section of dynamic expression one-to-one correspondence by manually observing the process and result of forming the dynamic expression on the epidermis; s4200, training to generate a mapping set composed of multiple groups of mapping, so that when a group of expression function vectors is input into the expression model, a corresponding dynamic expression is generated on the epidermis.
  9. 9. The method of claim 8, further comprising a reinforcement learning method of S5000, specifically comprising: S5100, the 4D expression model comprises a reciprocal 4D expression reverse model which corresponds to each other one by one, and a group of expression function vectors are obtained according to a segment of expression model in the mapping set; S5200, the 4D expression model further comprises a 4D shooting conversion model, and a section of video of the physical expression shot by 4D shooting is converted into a section of table condition model, wherein the 4D shooting specifically comprises front shooting and side shooting of 4D dynamic skin, and the front shooting and the side shooting are synthesized into a section of table condition model according to triangle synthesis; S5300, the 4D expression model further comprises an expression error model for calculating an error A between more than two sections of expression models, wherein the calculation of the error A comprises a block expansion error, a line expansion error and a 4D error of the 4D dynamic skin, and according to practical application, the error A does not exceed an allowable value; s5400 the 4D expression model further comprises an expression vector error model for calculating an error B between more than two groups of expression function vectors, wherein the calculation of the error B comprises the 4D errors of x, y, z and t in the expression function vectors, and i, p, v and tau errors, so that the error B does not exceed an allowable value according to practical application; s5500, respectively adjusting the block expansion error, the line expansion error and the 4D error of the 4D dynamic skin through a 4D camera shooting conversion model according to the expression error model, judging that the expression model and the expression function vector map at the moment are qualified when the error is smaller than an allowable value, and incorporating the map into a map set, otherwise judging that the map is unqualified, and/or, S5600, according to the expression vector error model, respectively adjusting vectors x, y, z, t and i, p, v and tau through a 4D shooting conversion model, calculating errors, and when the errors are smaller than the allowable values, judging that the expression model and the expression function vector mapping at the moment are qualified, and incorporating the mapping into a mapping set, otherwise judging that the mapping is unqualified.
  10. 10. The method according to claim 7 or 8 or 9, further comprising the following method of S6100 and S6200: S6100, the driving module further comprises a wireless communication sub-module, and the 4D expression model further comprises management and support of the wireless communication sub-module; S6200, wherein the special APP is used for generating communication with the wireless communication sub-module and the 4D expression model by scanning the two-dimensional code on the operation hardware except the 4D dynamic skin, the operation hardware comprises a computer, a mobile phone, a remote controller and the special hardware, the special APP also comprises functions of supervised learning and reinforcement learning in the 4D expression model, and a calculation result is submitted to the 4D expression model through the wireless communication sub-module so as to accelerate calculation force by using the operation hardware.

Description

Model construction method of VEM-4D dynamic skin structure and dynamic expression Technical Field The invention relates to the field of artificial intelligence, in particular to skin structure and control for a robot body and a robot animal, especially to the subdivision field of skin with dynamic expression for the robot body, aiming at realizing skin beat peristaltic motion. Background With the development of the robot industry, the robot has been widely used due to innovation of various moving parts of limbs of the robot. However, the current progress in the skin of the robot body is still not satisfactory, and in particular, the facial skin, which is a core component of emotion expression of the artificial intelligence robot body, has not been proposed until now, and thus, there is no suitable technology and manufacturing scheme, and in fact, this dilemma has prevented the development and application of the robot body as an anthropomorphic robot body. According to the search of the invention team, the following types of skin technical schemes exist at present: 1. Material driving type Pneumatic/hydraulic driving soft skin is to embed micro pneumatic network (pneumatic artificial muscle, PAM) inside elastic material such as silica gel. By controlling the air pressure of different cavities, the skin is locally expanded, contracted and bent to form smile, frowning and other expressions. The method has the advantages of soft action, high bionic degree, relatively slow response speed, complex pneumatic control system and air leakage risk. Shape memory material: shape Memory Alloy (SMA) wire/spring is embedded in skin base, and contracts after being electrified and heated, and pulls skin to deform. Its advantages are high strength and compactness, high power consumption, difficult thermal management, long service life and low response speed. Shape Memory Polymers (SMP) that can undergo preprogrammed macroscopic deformation by stimulation with heat, light, or the like. Many are used for global contour changes, but are more challenging to simulate locally, subtly, and quickly. Electroactive polymers (EAPs) such as Dielectric Elastomers (DE). Can generate great strain under the action of an electric field, and is similar to artificial muscle. The theoretical potential is great, but extremely high driving voltages (kilovolts) are required in practical applications, and material durability and packaging technology remain challenges. Tendon-rope drive, namely arranging bionic tendons (such as high-strength strings or hoses) under the flexible skin, and pulling the bionic tendons by a distal motor to realize the displacement of specific points of the skin. This is a classical approach to bionics, with relatively straightforward control, but with complex systems, with easy mechanical losses and noise generation. 2. Mechanical structure type A micro motor/linear driver array with multiple degrees of freedom is composed of miniature servo mechanism installed under the key points of face (such as eyebrow tip and mouth corner) for directly pushing or pulling skin. This is currently the dominant solution for many humanoid robots (e.g. Sophia, ameca). The advantages are accurate control, quick response, the disadvantages are that the face may appear mechanical, noisy, and the skin deformation in the non-driving point area may be unnatural. And the linkage link mechanism is a precise mechanical link, is driven by one or a few motors and realizes coordinated movement of a plurality of expression points through mechanism conversion. The control system can be simplified, but the mechanical design precision requirement is extremely high, the expression mode is fixed, and the flexibility is limited. 3. Hybrid drive and integration system The rigid-flexible hybrid drive is that a motor is used in the core area (such as opening and closing of the mouth), and the fine expression (such as cheek bulge and wrinkles) is driven by using soft bodies such as air power or EAP. Aims at balancing strength, speed and naturalness. Skin with multilayer composite structure: The functional lamination comprises an outermost layer which is a bionic texture and colored silica gel layer, a middle strain layer which is embedded with driving elements (such as SMA wires and fluid channels), and an inner layer which is possibly a basal layer connected with a robot face skeleton. And the sensing integration is to integrate flexible pressure, strain and even temperature sensors in the skin to form a 'sensing-driving' closed loop, so that the robot can sense touch and make more flexible reaction. VEM-Token model technique The team of the invention provides a group of novel innovation technologies of VEM (voice-Emotion-Multimodal, sound emotion multi-mode) for the first time, which comprises a patent pool consisting of the following 5 issued invention patents: 4.1 VEM-Token vocal emotion multi-modal model The model in the VEM-Token vocal emotion multimode To