A body-attached intelligent dexterous hand wireless sensing method and system

Abstract

The application discloses a kind of embodied intelligent dexterous hand wireless sensing method and system, for identifying the container containing target liquid from a group of identical transparent containers, belong to robot and embodied intelligent technical field.Combining robot visual identification and trained dielectric constant prediction model, the liquid dielectric-optical fingerprint library of liquid is established, so that the robot can judge whether the liquid type in the container is the target type according to the instruction.The application is suitable for family care, liquid transport and other scenes, and can also be widely used in industrial quality inspection, medical assistance, smart home and other application environments.Compared with traditional robots, the integrated design improves the identification efficiency and accuracy, meets the intelligent needs of real-time liquid property identification and transport and grasping in different scenes.

Description

Technical Field

[0001] This invention belongs to the field of robotics and embodied intelligence technology, and particularly relates to a wireless sensing method and system for embodied intelligent dexterous hand. Background Technology

[0002] In today's era of deep integration between intelligent manufacturing and the service industry, robotic systems are being given more diverse tasks. Liquid handling, as a common and critical operation, is widely used in many fields such as industrial production, medical care, and home services. Liquids have unique physical properties (such as fluidity, transparency, and volatility), which presents robots with many technical challenges when identifying, handling, and processing liquids.

[0003] Currently, liquid identification technology mainly relies on two categories of methods: contact and non-contact. Contact detection methods (such as electrochemical sensors and impedance detection) can provide relatively accurate liquid parameters, but they require direct immersion of the probe in the liquid, posing a risk of sample contamination and making them unsuitable for detecting sealed containers or hazardous chemicals. Furthermore, the probes of contact sensors are easily corroded and contaminated by liquids, affecting performance and lifespan, and increasing maintenance costs. Non-contact detection technologies (such as optical detection methods and ultrasonic technology) avoid some of the shortcomings of contact sensors, but they also have limitations. Optical detection methods (such as near-infrared spectroscopy) have high detection accuracy, but are easily interfered with in strong light or reflective environments, and their performance degrades in poor lighting conditions or when liquid surface features are not obvious. While ultrasonic technology has a certain penetrating ability, at high frequencies, sound waves attenuate rapidly in air, resulting in a short effective range and a decrease in resolution with increasing distance, making it difficult to achieve high-precision liquid property identification.

[0004] Meanwhile, existing liquid recognition and grasping systems often suffer from insufficient coordination between sensing and execution mechanisms. Data acquired by the sensing system cannot be fed back to the execution mechanism in real time and accurately, preventing the robot from adjusting its grasping strategy based on the real-time state of the liquid. This disconnect between perception and execution limits the robot's performance and efficiency in complex liquid handling tasks. Summary of the Invention

[0005] This invention addresses the shortcomings of current dexterous hand systems in liquid grasping tasks by proposing a wireless sensing method and system for an embodied intelligent dexterous hand, which can accurately identify and locate liquids inside transparent containers.

[0006] The technical solution adopted in this invention is as follows:

[0007] In a first aspect, the present invention provides a wireless sensing method for an embodied intelligent dexterous hand, used to identify containers containing a target liquid from a set of identical transparent containers, comprising the following steps:

[0008] (1) Design an intelligent dexterous hand, which is mounted on a robot with visual recognition capabilities. The fingers of the dexterous hand are integrated with a flexible ultra-wideband antenna array, which can emit electromagnetic waves of a specific frequency band to penetrate the liquid container under test and receive the returned penetration signal.

[0009] (2) The robot visually identifies the RGB values ​​of each type of liquid at different concentrations to obtain the first mapping relationship between different liquid types and RGB value ranges; the robot grasps the container filled with liquid with a dexterous hand, transmits electromagnetic wave signals by an antenna array and receives the returned transmission signals, and obtains the relative signal intensity and relative phase shift of the transmission signal based on the measurement method of electromagnetic wave transmission through liquid. The transmission signal, relative signal intensity, relative phase shift and RGB value range are used as liquid features, and the dielectric constant of the liquid is used as a training label to train a dielectric constant prediction model. The dielectric constant of different liquid types is generated using the trained dielectric constant prediction model to obtain the second mapping relationship between different liquid types and dielectric constants; the liquid dielectric-optical fingerprint database of liquids is established by combining the two mapping relationships.

[0010] (3) The robot receives user instructions through GUI or voice interaction, analyzes the type of target liquid to be grasped according to the instructions, locks the position of candidate container through robot vision analysis, controls the dexterous hand to grasp the candidate container and judges whether the type of liquid in the grasped container is the target type through dielectric constant prediction model. If it is, the container is retrieved to the designated position; otherwise, other candidate containers are replaced and identified in turn.

[0011] Furthermore, the transmission parameters of the flexible ultra-wideband antenna array include the center frequency and transmission power of the transmitted signal. These parameters are optimized by the return loss S11 of the transmitted signal itself and the return loss S11' of the returned transmitted signal. The transmission parameters with the largest difference between S11 and S11' are taken as the optimal transmission parameters.

[0012] Furthermore, the antenna array operates at a frequency of 2-5 GHz.

[0013] Furthermore, the process of obtaining the relative signal intensity and relative phase shift of the transmitted signal based on the measurement method of electromagnetic waves penetrating liquid includes:

[0014] Tests were conducted on empty containers and containers filled with different types of liquids. Electromagnetic waves were emitted and the returned transmission signals were received. The attenuation intensity of the transmission signal from the empty container was extracted. and phase shift ; and the attenuation intensity of the penetration signal in different types of liquid containers. and phase shift ; Calculate the relative signal strength between an empty cup container and different types of liquid containers. and relative phase shift ;in, , These represent the relative signal strength and relative phase shift of the penetration signal between the container containing the i-th type of liquid and the empty container, respectively.

[0015] Furthermore, the flexible ultra-wideband antenna array integrated on each of the dexterous fingers works independently to obtain the penetration signal, and the relative signal strength and relative phase shift of the penetration signal are obtained based on the measurement method of electromagnetic wave transmission through liquid.

[0016] Furthermore, it also includes a step of determining the liquid level in the container using a fusion of visual and electromagnetic wave sensing methods, including:

[0017] The overall height and liquid level of a transparent container are identified by a robot vision system, and the visual observation value of the liquid level is calculated.

[0018] Control the dexterous hand to grasp the candidate container, number the fingers from top to bottom according to their positions, and work synchronously with the antenna array to extract the relative signal strength and relative phase shift of the received signal at each finger position. Record the numbers of "empty" fingers with a relative signal strength close to 1 and a relative phase shift close to 0, and count the number of "empty" finger numbers and their distribution positions.

[0019] If the "empty" finger area inferred from the visual observation of the liquid level matches the distribution location of the "empty" finger numbers sensed by electromagnetic waves, then the final liquid level height will be based on the visual observation; otherwise, a re-detection or alarm will be triggered.

[0020] Furthermore, the polarization mode of the antenna array is selected according to the shape of the liquid container to be tested. The shape of the container is identified by robot vision. Circular cups are polarized vertically, while square cups are polarized horizontally.

[0021] Furthermore, the antenna array is equipped with a pressure sensor, which emits electromagnetic wave signals when the pressure value exceeds a threshold.

[0022] Furthermore, the dielectric constant prediction model employs either Gradient Boosting Decision Tree (XGBoost) or LightGBM.

[0023] Secondly, the present invention provides a wireless sensing system for an embodied intelligent dexterous hand, used to implement the aforementioned wireless sensing method for an embodied intelligent dexterous hand.

[0024] The beneficial effects of this invention are:

[0025] This invention provides a liquid recognition method and system based on the fusion of visual and electromagnetic wave perception. It integrates a flexible ultra-wideband antenna array into the dexterous finger, and achieves non-contact wireless perception of liquid information in a transparent container without affecting the normal grasping function of the dexterous hand.

[0026] In terms of sensing methods, this invention innovatively integrates visual optical information (RGB values) with electromagnetic wave sensing information (relative signal strength, relative phase shift). By establishing a liquid dielectric-optical fingerprint database that integrates the optical properties and dielectric constant of the liquid, and using machine learning algorithms for modeling, the accuracy and robustness of liquid recognition are significantly improved. This multimodal sensing approach effectively overcomes the limitations of traditional single-sensor modes and solves the problems of low recognition efficiency, poor accuracy, and susceptibility to environmental interference in traditional liquid recognition technologies.

[0027] Furthermore, this invention organically combines the identification and grasping processes, forming an automated closed-loop system from command reception, visual positioning, liquid identification to final grasping, thus improving the success rate and safety of grasping. This system fully utilizes the ability of vision to capture color information under good lighting conditions and the ability of electromagnetic waves to detect the internal dielectric properties of objects; the two complement each other. This system has broad application prospects in smart homes, laboratory automation, and retail services, providing a reliable technical solution for achieving accurate liquid identification and grasping operations. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating the wireless sensing method of the embodied intelligent dexterous hand in this invention;

[0029] Figure 2 This is a schematic diagram illustrating liquid recognition in a water cup within a smart home scenario;

[0030] Figure 3 This is a schematic diagram of the framework of the embodied intelligent dexterous hand wireless sensing system in this invention. Detailed Implementation

[0031] The present invention will be further described and illustrated below with reference to specific embodiments. The embodiments described are merely examples of the content of this disclosure and do not limit the scope of the invention. The technical features of each embodiment in the present invention can be combined accordingly, provided that there is no mutual conflict.

[0032] like Figure 1 As shown, this invention proposes a wireless sensing method for an embodied intelligent dexterous hand, used to identify containers containing a target liquid from a group of identical transparent containers, mainly including the following steps:

[0033] S1 designs an embodied intelligent dexterous hand for use in visual recognition robots, whose fingers integrate a flexible ultra-wideband antenna array that can emit and receive electromagnetic waves of specific frequencies to penetrate and sense liquids inside containers.

[0034] In this step, an intelligent dexterous hand is designed, which is mounted on a robot with visual recognition capabilities. The fingers of the dexterous hand integrate a flexible ultra-wideband antenna array, which can emit electromagnetic waves of a specific frequency band to penetrate the liquid container under test and receive the returned penetration signal.

[0035] S2 trains a dielectric constant prediction model by fusing visual RGB information with electromagnetic wave signal features (relative intensity and phase shift), thereby constructing a liquid dielectric-optical fingerprint library that combines optical and electrical properties.

[0036] In this step, the robot's vision identifies the RGB values ​​of each type of liquid at different concentrations, obtaining a first mapping relationship between different liquid types and RGB value ranges. A dexterous hand grasps a container filled with liquid, and an antenna array emits electromagnetic wave signals and receives the returned transmission signals. Based on the measurement method of electromagnetic wave transmission through liquid, the relative signal intensity and relative phase shift of the transmission signals are obtained. Using the transmission signal, relative signal intensity, relative phase shift, and RGB value range as liquid features, and the dielectric constant of the liquid as a training label, a dielectric constant prediction model is trained. The trained dielectric constant prediction model is then used to generate the dielectric constants of different liquid types, obtaining a second mapping relationship between different liquid types and dielectric constants. Finally, a liquid dielectric-optical fingerprint database of liquids is established by combining the two mapping relationships.

[0037] S3: The robot receives instructions via voice or GUI, locates the container visually, controls its dexterous hand to grasp it, and uses a dielectric constant model to determine whether the liquid is the target type, thereby deciding whether to retrieve or replace the container.

[0038] In this step, the robot receives user instructions through GUI or voice interaction, analyzes the type of target liquid to be grasped according to the instructions, locks the position of candidate containers through robot vision analysis, controls the dexterous hand to grasp the candidate containers, and uses the dielectric constant prediction model to determine whether the type of liquid in the grasped container is the target type. If it is, the container is retrieved to the designated position; otherwise, other candidate containers are used for identification in sequence.

[0039] In one specific embodiment of the present invention, when employing a measurement method based on electromagnetic wave transmission through liquid, it is necessary to first determine the transmission parameters of the flexible ultra-wideband antenna array. These parameters include the center frequency and transmission power of the transmitted signal. Optimization is achieved through the return loss S11 of the transmitted signal itself and the return loss S11' of the returned transmitted signal. The transmission parameters with the largest difference between S11 and S11' are taken as the optimal transmission parameters. This ensures that more electromagnetic wave energy penetrates the container wall and enters the liquid, rather than being reflected or dissipated, thus enhancing the sensitivity to changes in the liquid. Preferably, the antenna array is equipped with a pressure sensor. When the pressure value exceeds a threshold, an electromagnetic wave signal is transmitted. This ensures that during electromagnetic wave transmission, the propagation loss of the electromagnetic wave in the air and the reflection loss at the container wall are greatly reduced, making the distance between the antenna and the outer wall of the container negligible.

[0040] Generally, the operating frequency of the antenna array is 2-5GHz. The polarization mode of the antenna array is selected according to the shape of the liquid container to be tested. The shape of the container is identified by robot vision. Circular cups use vertical polarization, and square cups use horizontal polarization. Different polarization modes are used to optimize the penetration and reception efficiency of electromagnetic waves.

[0041] In one specific embodiment of the present invention, the process of establishing visual recognition and RGB mapping in step S2 is as follows:

[0042] The robot vision system identifies the RGB values ​​of each liquid at different concentrations. For example, it collects multiple images of different types of liquids (such as water, ethanol, gasoline, juice, etc.) at different concentrations; uses the OpenCV image processing library to extract the average RGB values ​​of the liquids in the images; and calculates the RGB value range (R_min, R_max, G_min, G_max, B_min, B_max) of each liquid at different concentrations, establishes the first mapping relationship between different liquid types and RGB value ranges, and stores the data.

[0043] In step S2, the process of obtaining the relative signal intensity and relative phase shift of the penetration signal based on the measurement method of electromagnetic wave transmission through liquid includes:

[0044] Tests were conducted on empty containers and containers filled with different types of liquids. Electromagnetic waves were emitted and the returned transmission signals were received. Optionally, based on signal differential processing and adaptive filtering algorithms, filtering parameters were dynamically adjusted to suppress interference and background noise in the transmission signals. The attenuation intensity of the transmission signal from the empty container was extracted. and phase shift ; and the attenuation intensity of the penetration signal in different types of liquid containers. and phase shift ; Calculate the relative signal strength between an empty cup container and different types of liquid containers. and relative phase shift ;in, , These represent the relative signal strength and relative phase shift of the penetration signal between the container containing the i-th type of liquid and the empty container, respectively.

[0045] Here, the flexible ultra-wideband antenna array integrated on each of the dexterous fingers works independently to obtain the penetration signal, and the relative signal strength and relative phase shift of the penetration signal are obtained based on the measurement method of electromagnetic waves penetrating liquid.

[0046] In one specific embodiment of the present invention, the dielectric constant prediction model employs either Gradient Boosting Decision Tree (XGBoost) or LightGBM. The process of training the dielectric constant prediction model includes: collecting a large amount of sample data, including liquid characteristics (penetration signal, relative signal strength, relative phase shift, RGB value range) as input features, and known liquid dielectric constants as training labels; dividing the dataset into training, validation, and test sets; training the model using the training set and adjusting hyperparameters; evaluating model performance using the validation set with metrics such as accuracy and mean squared error, and selecting a model; evaluating the performance of the final model using the test set; and using the trained model to predict the dielectric constant of new liquids, establishing a second mapping relationship between liquid types and dielectric constants.

[0047] By combining the first mapping relationship (liquid type - RGB value range) and the second mapping relationship (liquid type - dielectric constant), a liquid dielectric-optical fingerprint database containing liquid type, RGB value range, and dielectric constant is constructed. This fingerprint database can be used as prior knowledge for subsequent liquid identification.

[0048] This invention also implements a step for determining the liquid level in a container based on a fusion of visual and electromagnetic wave sensing, including:

[0049] The robot vision system identifies the overall height and liquid level position of a transparent container, and calculates the visual observation value of the liquid level. During detection, the three-dimensional contour of the container opening and the edge of the liquid surface can be extracted in real time based on the U-Net architecture to identify the liquid height, with a positioning error of ≤±0.1 mm. Visual measurement can usually provide high-precision liquid level information when the lighting is good and the liquid surface features are obvious, but it is easily affected by factors such as ambient light, liquid surface shape, container wall refraction and camera angle.

[0050] The dexterous hand grasps candidate containers, numbering fingers from top to bottom based on their position. An antenna array operates synchronously, extracting the relative signal strength and phase shift of the received signal at each finger position. "Empty" fingers with relative signal strength close to 1 and relative phase shift close to 0 are recorded, and their number and distribution are counted. The principle behind this step is as follows: When the dexterous hand grasps the container, the antenna array integrated on the fingers emits electromagnetic waves and receives signals that penetrate the container wall and liquid. By analyzing the relative signal strength and relative phase shift, it determines whether the corresponding position is liquid or air. If the relative signal strength is close to 1 and the relative phase shift is close to 0, it indicates that the signal characteristics of that finger position are very similar to the "empty cup" reference state, therefore, it is determined that there is no liquid in that area and it is marked as an "empty" finger. The antennas on multiple fingers operate independently, simultaneously acquiring liquid level information at different heights within the container, providing a one-dimensional overview of the liquid level distribution.

[0051] If the "empty" finger area inferred from the visual observation of the liquid level matches the distribution of the "empty" finger numbers sensed by electromagnetic waves, then the final liquid level height adopts the visual observation value; otherwise, a re-detection or alarm is triggered to prevent the robot from performing subsequent operations based on unreliable information, thereby improving the functional safety level of the entire system.

[0052] Here, vision provides an intuitive liquid level value, while electromagnetic wave sensing provides stable but coarse evidence of the liquid level distribution. The two sensors cross-check each other, greatly reducing the risk of system errors caused by misjudgment from a single sensor.

[0053] Figure 2 This is a schematic diagram of liquid recognition in a smart home scenario. Taking a water dispensing command as an example, one possible liquid recognition and grasping process is as follows:

[0054] The robot receives user commands (such as "get a bottle of water") via GUI or voice interaction, analyzes the type of liquid to be grasped, and analyzes the environment to identify and locate candidate containers. It then controls its dexterous hand to grasp the candidate container (if two containers are close together, it can grasp both simultaneously for identification), and its antenna array emits electromagnetic wave signals and receives the returned signals. The robot extracts the signal features (relative signal strength, relative phase shift) and visual features (RGB values) of the current container. These features are input into a trained dielectric constant prediction model to obtain the predicted dielectric constant. The robot then queries a liquid dielectric-optical fingerprint database to determine if the current liquid matches the target liquid. If so, the dexterous hand retrieves the container to a designated location; otherwise, the robot releases the current container and identifies other candidate containers sequentially. In one scenario, the robot's vision system identifies three transparent cups (A, B, C) on a table. The dexterous hand first grasps cup A, the antenna emits electromagnetic waves, the signal processing unit extracts features, the model predicts its dielectric constant is close to that of water, the fingerprint database confirms it is water, and the robot retrieves cup A and hands it to the user. If cup A is identified as another liquid, then cups B and C are identified in turn.

[0055] The present invention also provides a wireless sensing system for an embodied intelligent dexterous hand, used to implement the above-mentioned wireless sensing method for an embodied intelligent dexterous hand, the system comprising:

[0056] The robot body integrates a self-contained intelligent dexterous hand and a visual recognition system. The fingers of the dexterous hand integrate a flexible ultra-wideband antenna array, which can emit electromagnetic waves of a specific frequency band to penetrate the liquid container under test and receive the returned penetration signal.

[0057] The dielectric constant prediction model module is used to predict the dielectric constant based on the transmission signal, relative signal strength, relative phase shift, and RGB value range. The transmission signal is obtained by transmitting electromagnetic wave signals through an antenna array and receiving the returned signals when a dexterous hand grasps a container filled with liquid. The relative signal strength and relative phase shift are obtained by measuring the transmission of electromagnetic waves through the liquid. The RGB value range is obtained by identifying the RGB values ​​of each type of liquid at different concentrations through a visual recognition system.

[0058] A liquid dielectric-optical fingerprint library is used to store the mapping relationship between different liquid types, RGB value ranges, and dielectric constants.

[0059] The interaction module is used to receive user commands through GUI or voice interaction, analyze the type of target liquid to be grasped according to the command, control the robot to lock the position of the candidate container through the vision recognition system, control the dexterous hand to grasp the candidate container, and use the dielectric constant prediction model to determine whether the type of liquid in the grasped container is the target type. If it is, the container is retrieved to the designated position; otherwise, other candidate containers are identified in turn.

[0060] In the above system embodiment, the robot body can be a commercial robot platform with mobility and visual recognition capabilities. The dexterous hand integrated into the robot body is a five-fingered dexterous hand, with each finger wrapped in flexible material and a flexible ultra-wideband antenna and pressure sensor integrated inside the fingertip. The visual recognition system integrated into the robot body uses a USB high-definition camera mounted on the robot's head for image acquisition. The dielectric constant prediction model module and the liquid dielectric-optical fingerprint library can be implemented using a Raspberry Pi 4B and an STM32F407 chip.

[0061] Building upon this, those skilled in the art can introduce a motion planning unit and a spill prevention module. The motion planning unit generates a grasping trajectory based on an adaptive algorithm and adjusts it to prevent liquid spillage during grasping. The spill prevention module calculates the liquid center of gravity offset ΔG to optimize the motion trajectory.

[0062] For example, one possible implementation of the spill prevention module is as follows:

[0063] Calculate the liquid center of gravity offset ΔG based on the container shape and liquid height (liquid level);

[0064] By combining data from the dexterity hand accelerometer, and using the liquid center of gravity offset ΔG and the dexterity hand acceleration as inputs, the dynamic sloshing trend of the liquid is predicted, and the sloshing amplitude changes over time is obtained.

[0065] A fuzzy PID control algorithm is used to adjust the grasping force and end motion trajectory in real time to ensure that the fluctuation range of the liquid surface is ≤5mm.

[0066] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. Those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A wireless sensing method for an embodied intelligent dexterous hand, used to identify containers containing a target liquid from a set of identical transparent containers, characterized in that, Includes the following steps: (1) Design an intelligent dexterous hand, which is mounted on a robot with visual recognition capabilities. The fingers of the dexterous hand are integrated with a flexible ultra-wideband antenna array, which can emit electromagnetic waves of a specific frequency band to penetrate the liquid container under test and receive the returned penetration signal. (2) The robot visually identifies the RGB values ​​of each type of liquid at different concentrations to obtain the first mapping relationship between different liquid types and RGB value ranges; the robot grasps the container filled with liquid with a dexterous hand, transmits electromagnetic wave signals by an antenna array and receives the returned transmission signals, and obtains the relative signal intensity and relative phase shift of the transmission signal based on the measurement method of electromagnetic wave transmission through liquid. The transmission signal, relative signal intensity, relative phase shift and RGB value range are used as liquid features, and the dielectric constant of the liquid is used as a training label to train a dielectric constant prediction model. The dielectric constant of different liquid types is generated using the trained dielectric constant prediction model to obtain the second mapping relationship between different liquid types and dielectric constants; the liquid dielectric-optical fingerprint database of liquids is established by combining the two mapping relationships. (3) The robot receives user instructions through GUI or voice interaction, analyzes the type of target liquid to be grasped according to the instructions, locks the position of candidate container through robot vision analysis, controls the dexterous hand to grasp the candidate container and judges whether the type of liquid in the grasped container is the target type through dielectric constant prediction model. If it is, the container is retrieved to the designated position; otherwise, other candidate containers are replaced and identified in turn.

2. The wireless sensing method for embodied intelligent dexterous hand according to claim 1, characterized in that, The transmission parameters of the flexible ultra-wideband antenna array include the center frequency and transmission power of the transmitted signal. They are optimized by the return loss S11 of the transmitted signal itself and the return loss S11' of the returned transmitted signal. The transmission parameters with the largest difference between S11 and S11' are taken as the optimal transmission parameters.

3. The wireless sensing method for embodied intelligent dexterous hand according to claim 1, characterized in that, The antenna array operates at a frequency of 2-5 GHz.

4. The wireless sensing method for embodied intelligent dexterous hand according to claim 1, characterized in that, The process of obtaining the relative signal intensity and relative phase shift of the transmitted signal based on the measurement method of electromagnetic waves through liquid includes: Tests were conducted on empty containers and containers filled with different types of liquids. Electromagnetic waves were emitted and the returned transmission signals were received. The attenuation intensity of the transmission signal from the empty container was extracted. and phase shift ; and the attenuation intensity of the penetration signal in different types of liquid containers. and phase shift ; Calculate the relative signal strength between an empty cup container and different types of liquid containers. and relative phase shift ;in, , These represent the relative signal strength and relative phase shift of the penetration signal between the container containing the i-th type of liquid and the empty container, respectively.

5. The wireless sensing method for embodied intelligent dexterous hand according to claim 4, characterized in that, Each flexible ultrawideband antenna array integrated on a dexterous finger works independently to obtain the penetration signal, and obtains the relative signal strength and relative phase shift of the penetration signal based on the measurement method of electromagnetic waves penetrating liquid.

6. The wireless sensing method for embodied intelligent dexterous hand according to claim 5, characterized in that, It also includes a step of determining the liquid level in the container using a fusion of visual and electromagnetic wave sensing methods, including: The overall height and liquid level of a transparent container are identified by a robot vision system, and the visual observation value of the liquid level is calculated. Control the dexterous hand to grasp the candidate container, number the fingers from top to bottom according to their positions, and work synchronously with the antenna array to extract the relative signal strength and relative phase shift of the received signal at each finger position. Record the numbers of "empty" fingers with relative signal strength close to 1 and relative phase shift close to 0, and count the number of "empty" finger numbers and their distribution positions. If the "empty" finger area inferred from the visual observation of the liquid level matches the distribution location of the "empty" finger numbers sensed by electromagnetic waves, then the final liquid level height will be based on the visual observation; otherwise, a re-detection or alarm will be triggered.

7. The wireless sensing method for embodied intelligent dexterous hand according to claim 1, characterized in that, The polarization of the antenna array is selected based on the shape of the liquid container being tested. The shape of the container is identified by robot vision. Circular cups are polarized vertically, while square cups are polarized horizontally.

8. The wireless sensing method for embodied intelligent dexterous hand according to claim 1, characterized in that, The antenna array is equipped with a pressure sensor, which emits electromagnetic wave signals when the pressure value exceeds a threshold.

9. The wireless sensing method for embodied intelligent dexterous hand according to claim 1, characterized in that, The dielectric constant prediction model uses either Gradient Boosting Decision Tree (XGBoost) or LightGBM.

10. A wireless sensing system for an embodied intelligent dexterous hand, used to implement the wireless sensing method for an embodied intelligent dexterous hand as described in claim 1, characterized in that, The system includes: The robot body integrates a self-contained intelligent dexterous hand and a visual recognition system. The fingers of the dexterous hand integrate a flexible ultra-wideband antenna array, which can emit electromagnetic waves of a specific frequency band to penetrate the liquid container under test and receive the returned penetration signal. The dielectric constant prediction model module is used to predict the dielectric constant based on the transmission signal, relative signal strength, relative phase shift, and RGB value range. The transmission signal is obtained by transmitting electromagnetic wave signals through an antenna array and receiving the returned signals when a dexterous hand grasps a container filled with liquid. The relative signal strength and relative phase shift are obtained by measuring the transmission of electromagnetic waves through the liquid. The RGB value range is obtained by identifying the RGB values ​​of each type of liquid at different concentrations through a visual recognition system. A liquid dielectric-optical fingerprint library is used to store the mapping relationship between different liquid types, RGB value ranges, and dielectric constants. The interaction module is used to receive user commands through GUI or voice interaction, analyze the type of target liquid to be grasped according to the command, control the robot to lock the position of the candidate container through the vision recognition system, control the dexterous hand to grasp the candidate container, and use the dielectric constant prediction model to determine whether the type of liquid in the grasped container is the target type. If it is, the container is retrieved to the designated position; otherwise, other candidate containers are identified in turn.

Images