A Tai Chi Pushing Hands Robotic Arm Based on Electronic Skin

Through a modular electronic skin sensing system and deep reinforcement learning, the problem of traditional robotic arms' difficulty in accessing information in human-computer interaction is solved, and the accurate coordination and safety of robotic arms in activities such as Tai Chi pushers is achieved.

CN115723175BActive Publication Date: 2024-12-03TONGJI UNIV
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202211428193.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-11-15
Publication Date
2024-12-03
Estimated Expiration
2042-11-15

AI Technical Summary

Technical Problem

Traditional robotic arms are easily disturbed by the environment through visual acquisition of external information, and cannot directly obtain contact information during human-computer interaction, and the torque sensor makes it bulky and unable to accurately fit the magnitude and position of the combined force.

Method used

The electronic skin feeling system with a modular design is adopted, combined with the central decision-making system and control execution system, and through multi-sensor fusion algorithm and deep reinforcement learning, the extraction and control strategy of contact mode is realized, and the robot arm is driven to perform corresponding actions.

Benefits of technology

Electronic skins can quickly obtain contact information, assist robotic arms to judge human intentions, improve the accuracy and safety of human-computer interaction, especially the ability to cooperate with humans in activities such as Tai Chi pushers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115723175B_ABST
    Figure CN115723175B_ABST
Patent Text Reader

Abstract

The present invention discloses a Tai Chi Push Hands robot arm based on electronic skin, which at least comprises a sensing unit, an execution unit, a control unit and a decision unit. The sensing unit is a modular electronic skin that can be covered on the surface of the robot arm, and can sense various physical quantities such as contact force, proximity, temperature, acceleration, etc.; the execution unit is a motor in the robot arm, and can drive the robot arm to complete corresponding Tai Chi Push Hands actions according to an upper-level control signal; the control unit generates a corresponding control signal based on the reasoning and analysis of the decision unit, and sends it to the execution unit; the decision unit has a built-in tactile perception fusion algorithm, which utilizes a multi-sensor fusion algorithm to fuse the data received by the electronic skin, and infers the movement trajectory and intention of the human arm when performing Tai Chi massage. The present invention can realize that the robot arm and humans work closely together to complete a series of Tai Chi massage actions, and meet the needs of Tai Chi Push Hands for specific groups of people under the premise of achieving safe interaction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of sensing and control technology, and in particular to a Tai Chi push-hands mechanical arm based on electronic skin. Background Art

[0002] Human-machine interaction is one of the major trends in the future. When humans and robots work together to complete specific tasks, robots must not only be able to perceive the external environment and make decisions, but also understand human intentions to avoid harming humans. With the development of technology, intelligent service robots or robotic arms have become a new research hotspot. If robotic arms can have a sense of touch similar to that of humans, then the intelligent perception of robotic arms will take an important step forward. Therefore, the electronic skin wrapped around the robot has become a key component with high application value. The control and decision-making of service robotic arms based on electronic skin also have important research significance.

[0003] Traditionally, robots or robotic arms mainly obtain external information through vision, which is easily affected by environmental interference, such as object occlusion, camera blind spots and other factors. In addition, the direct contact information generated during human-computer interaction cannot be directly obtained through the captured video images. If torque sensors are added to the joints of the robotic arm, the robot or robotic arm will become cumbersome and unable to fit the accurate size and position of the resultant force. Summary of the invention

[0004] The purpose of the present invention is to solve the above problems and provide a Tai Chi Push Hands robot arm based on electronic skin, comprising a robot arm body, an electronic skin sensing system wrapped in the robot arm body, a central decision system connected to the electronic skin sensing system, and a control execution system connected to the central decision system signal;

[0005] The electronic skin sensing system is modularly designed, including multiple sensing modules. Each sensing module includes a front-end sensor element, a data processing unit and a data transmission unit. The data processing unit receives and processes the information of the front-end sensor element and transmits the information to the central decision-making system through the data transmission unit.

[0006] The central decision-making system includes a contact pattern extraction unit and a decision-making unit. The contact pattern extraction unit receives the signal transmitted by the data transmission unit based on a multi-sensor fusion algorithm and extracts the contact semantics. The decision-making unit judges human intentions based on deep reinforcement learning and generates a control strategy.

[0007] The control execution system includes a control unit and an execution unit that are interconnected. The control unit receives the control strategy, generates a corresponding control signal, and transmits the control signal to the execution unit. The execution unit includes a motor, which drives the robotic arm to complete the corresponding action through the control signal.

[0008] Furthermore, the front-end sensing element is one or more combinations of a pressure sensor, a proximity sensor, a vibration sensor, a temperature sensor, an acceleration sensor, a magnetic sensor, a stretch sensor, a gyroscope and a magnetometer.

[0009] Furthermore, the modular electronic skin sensing system is cut or spliced ​​into a whole piece of electronic skin of any shape, and the whole piece of electronic skin is covered on the surface of the robotic arm with different shapes and curvatures. The robotic arm identifies and collects various physical quantities of the objects it contacts through the electronic skin, and processes the various physical quantities through the data processing unit.

[0010] Furthermore, the data processing unit is an MCU, and the data processing unit has a built-in data processing algorithm. The data processing unit performs data processing and adaptive filtering and denoising through the data processing algorithm, and the electronic skin sensing system switches the algorithm automatically according to actual needs.

[0011] Furthermore, the data transmission unit is a wired transmitter or a wireless transmitter. When the data transmission unit is a wired transmitter, the wired transmitter is provided with a plurality of ports, and adjacent electronic skins and the electronic skin and the contact pattern extraction unit are connected to each other through the ports and the lines.

[0012] When the data transmission unit is a wireless transmitter, adjacent electronic skins and the electronic skin and the contact pattern extraction unit are connected via Bluetooth or WiFi signals.

[0013] Furthermore, the multi-sensor fusion algorithm includes the following steps:

[0014] S1: Decouple multiple targets and associate sensor information of the same event, cluster and classify different target data, obtain the contact time of the left and right arms of the robot, and then associate the data packets of the corresponding areas. When clustering and classifying different data, the cluster classification with too few activated units of the data is discarded;

[0015] S2: Data alignment, performing spatiotemporal matching based on the position and time information in the data packet, unifying the position and time information into the robot global coordinate system and the global unified time node;

[0016] S3: The pressure magnitude and direction when the human hand contacts the robotic arm are obtained based on the values ​​of the three-dimensional pressure sensor, and the total force and torque magnitude are calculated based on the orientation of the values ​​in the global coordinate system of the robotic arm;

[0017] S4: Kalman filtering is performed on the contact force and torque to predict the motion trajectory of the human arm;

[0018] S5: The prediction of the human arm motion trajectory and the magnitude of the contact force are put into the trained deep neural network. The output of the network is the predefined contact pattern, and finally the corresponding contact pattern is obtained.

[0019] Furthermore, in S5, the process of deep neural network training includes the following steps:

[0020] S51: Adjust the robotic arm to the teaching mode or the follow-up control mode, collect the raw data of all sensors when the robotic arm and the human perform normal Tai Chi push hands activities, align the group data in time and space, and obtain the original data set; initialize and construct a deep neural network, and set the initial parameters of the deep neural network. The structure of the deep neural network can use a conventional network structure, the number of input layer neurons is set according to the selected sensor, and the number of output layer neurons is set according to the electronic skin state quantity;

[0021] S52: performing preprocessing, feature extraction and standardization calculation on the measurement data of the sensor in the original data set to obtain a training data set; the standardization calculation includes performing Kalman filtering on the data to obtain a motion trajectory;

[0022] S53: Use the training data set to train the deep neural network to obtain a trained deep neural network.

[0023] Furthermore, the decision-making unit uses a deep reinforcement learning method based on value learning to train the deep Q network to obtain the optimal action value function, including the following steps:

[0024] SA1: Set a reward function, which is related to the contact pattern and contact pressure, and the degree of consistency between the contact pressure and contact pattern and the preset trajectory and pattern is proportional to the score;

[0025] SA2: Training deep Q-network using temporal difference algorithm;

[0026] SA3: After training, the network output is the optimal action for the next step, which is the output of the decision unit.

[0027] Furthermore, a threshold is set for the degree of fit. When it is higher than the threshold, the contact pattern extracted by the contact pattern extraction unit is transmitted to the deep Q network in the decision unit; when it is lower than the threshold, the robotic arm issues a reminder and enters the compliant mode. The compliant mode is characterized by the robotic arm moving completely according to the trajectory of the human arm.

[0028] Furthermore, after the robotic arm enters the compliant mode, during any interaction process, when the electronic skin senses excessive external force, the robotic arm stops moving and issues a reminder. After a delay of 5 seconds, the robotic arm automatically returns to the initial position of Tai Chi Push Hands.

[0029] Compared with the prior art, the beneficial effects of the present invention are: the electronic skin can easily and quickly obtain contact information such as the contact direction and magnitude direction of the external force, and can better assist robots or robotic arms in judging contact patterns and human intentions. It has great advantages in scenarios such as human-computer interaction where humans and robotic arms have contact. Combined with the central decision-making system in the robotic arm, the robotic arm can cooperate with humans in activities based on compliant control, such as cooperating with humans to complete the fitness activity of Tai Chi Push Hands. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a schematic diagram of the operating principle of a Tai Chi Push Hands robotic arm based on electronic skin in the present invention. DETAILED DESCRIPTION

[0031] The following will provide a more detailed description of a Tai Chi Push Hands robotic arm based on electronic skin of the present invention in conjunction with a schematic diagram, in which a preferred embodiment of the present invention is shown. It should be understood that those skilled in the art can modify the present invention described herein and still achieve the beneficial effects of the present invention. Therefore, the following description should be understood as being widely known to those skilled in the art and not as a limitation to the present invention.

[0032] In the description of the present invention, it should be noted that directional words, such as the terms "center", "lateral", "longitudinal", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", etc., indicating directions and positional relationships are based on the directions or positional relationships shown in the accompanying drawings, which are only for the convenience of narrating the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and cannot be understood as limiting the specific scope of protection of the present invention.

[0033] A Tai Chi Push Hands robotic arm based on electronic skin is dedicated to wrapping an electronic skin sensing system on the outer surface of the robotic arm, which can realize sensing functions similar to human skin, including touch, pressure, vibration, pain, proximity and temperature, etc., so that the robotic arm can have a sense of touch similar to that of humans. Combined with the control execution and central decision-making system, this robotic arm can interact with humans. In the Tai Chi Push Hands scenario, it can infer the human arm movements and trajectories, judge human intentions, and cooperate with humans in Tai Chi Push Hands human-computer interaction activities.

[0034] like Figure 1As shown, a Tai Chi push-hands robot arm based on electronic skin needs to include at least a robot body, an electronic skin sensing system wrapped around the outside of the robot arm, a control execution system and a central decision-making system; wherein the electronic skin sensing system is modularly designed, and each module includes a front-end sensing element, a data processing unit and a data transmission unit, and the front-end sensing element includes but is not limited to one or more of a pressure sensor, a proximity sensor, a vibration sensor, a temperature sensor, an acceleration sensor, a magnetic sensor, a stretch sensor, a gyroscope and a magnetometer, etc., which can sense contact force, proximity, temperature, acceleration and other physical quantities; in this embodiment, the number of temperature sensors is multiple, which can monitor the temperature of external contact objects and detect the internal temperature of the electronic skin, use the temperature field to calculate the temperature of other sensors, and then use the temperature information to calibrate other sensors, so as to perform temperature compensation on sensors at different positions on the electronic skin, eliminating the influence of temperature on sensor measurement data.

[0035] The data processing unit is an MCU, which has built-in data processing algorithms such as Kalman filtering, median filtering, IIR filtering, etc. The system can switch algorithms according to actual needs.

[0036] The data transmission unit can adopt an external interface wired transmission method, or a built-in wireless transmitter wireless transmission method. Wired transmission can use I2C, SPI, Ethernet, CAN bus and other methods, and wireless transmission can use WIFI or Bluetooth. Multiple adjacent electronic skins can also communicate with each other. For example, an electronic skin P is connected to the host computer through a line, and the adjacent electronic skin transmits data to the electronic skin P, and then transmits it to the host computer via the line on the electronic skin P, which reduces the complexity of wiring. On the other hand, communication between adjacent electronic skins can correct data with each other, making the function of the electronic skin system more complete. The various modules can communicate with each other by wire or wireless connection. The communication methods include I2C, SPI, CAN, Ethernet, Bluetooth, WIFI and other methods. The communication method can be manually selected or the algorithm can automatically select the optimal transmission strategy.

[0037] Modular electronic skin can be cut or spliced ​​into a whole piece of electronic skin of any shape, so as to cover various irregular surfaces more completely and conformally. It can also use the electronic skin sensing system to sense the point and direction of external force, the temperature of the contact object, the impact force of the contact, the softness of the contact object and other physical quantities. After adaptive filtering and denoising by the data processing unit, the data is transmitted to the central decision-making system by wired or wireless means using the data transmission unit.

[0038] The central decision-making system mainly consists of two parts: the contact pattern extraction unit based on the multi-sensor fusion algorithm and the decision-making unit based on deep reinforcement learning. The contact pattern extraction unit is mainly responsible for the extraction of contact semantics, and the main steps are as follows:

[0039] 1): Decouple multiple targets and associate sensor information of the same event. Mainly cluster and classify different target data. If there are too few activated units in the cluster, this classification will be discarded. Finally, the contact event between the left arm and the right arm is obtained, and then the data packets of the corresponding areas are associated.

[0040] 2): Data alignment, mainly based on the time and space matching of the position and time information in the data packet. The installation positions of modular electronic skins are different, and the time when the stimulus arrives at the host computer at the same time is also different. It should be unified to the global coordinate system of the robot arm and the global unified time node.

[0041] 3): Based on the values ​​of the three-dimensional pressure sensor, the pressure magnitude and direction when the human hand contacts the robotic arm are obtained, and the total force and torque magnitude are calculated based on the orientation of the values ​​in the global coordinate system of the robotic arm.

[0042] 4): Perform Kalman filtering on the contact force and torque to predict the motion trajectory of the human arm.

[0043] 5): The prediction of the human arm's motion trajectory and the magnitude of the contact force are put into the trained deep neural network. The output of the network is the predefined contact pattern, and finally the corresponding contact pattern is obtained.

[0044] Among them, the training process of deep neural network includes the following steps:

[0045] S51: Adjust the robotic arm to the teaching mode or the follow-up control mode, collect the raw data of all sensors when the robotic arm and the human perform normal Tai Chi push hands activities, align the group data in time and space, and obtain the original data set; initialize and construct a deep neural network, and set the initial parameters of the deep neural network. The structure of the deep neural network can use a conventional network structure, the number of input layer neurons is set according to the selected sensor, and the number of output layer neurons is set according to the electronic skin state quantity;

[0046] S52: performing preprocessing, feature extraction and standardization calculation on the measurement data of the sensor in the original data set to obtain a training data set; the standardization calculation includes performing Kalman filtering on the data to obtain a motion trajectory;

[0047] S53: Use the training data set to train the deep neural network to obtain a trained deep neural network.

[0048] The decision-making unit is mainly based on deep reinforcement learning. The main training and use steps are as follows:

[0049] 1): Set a reward function, which is related to the contact mode and contact pressure. The more the contact pressure and contact mode match our preset trajectory and mode, the higher the score.

[0050] 2): Use the temporal difference algorithm (TDlearning) to train the deep Q network (DQN).

[0051] 3): After training is completed, the network output is the optimal action for the next step, which is the output of the decision unit.

[0052] The execution unit of the control execution system is the motor in the robotic arm, and the control unit is the joint controller. The execution unit can use the driver to drive the motor through the joint controller according to the upper control signal. The motor is connected to the transmission structure to drive the robotic arm to complete the corresponding Tai Chi pushing hands action.

[0053] In addition, it should be noted that:

[0054] In step 1) of the decision unit, after the robot arm infers the motion trajectory of the human arm, it will perform a similarity analysis between this trajectory and the real Tai Chi massage trajectory. If the similarity is lower than the previously set threshold, we set the threshold to 75%. Then a reminder will be issued and the robot arm will enter the compliant control mode, that is, the robot arm moves completely according to the trajectory of the human arm; if the predicted trajectory is similar to the real Tai Chi massage trajectory, that is, higher than the threshold, then the contact pattern extracted by the contact pattern extraction unit will be passed to the deep Q network in the decision unit.

[0055] During any interaction, whether in the normal mode or the compliant control mode of the robot arm, when the electronic skin senses an inappropriate or excessive external force, specifically: a force exceeding 20N is an inappropriate or excessive external force, the robot arm will stop moving, issue a reminder, and automatically return to the initial Tai Chi push hands preparation position after a delay of 5 seconds, thereby protecting humans and embodying interactive safety.

[0056] The overall signal transmission process of the present invention is as follows:

[0057] Each module of the electronic skin sensing system senses external stimuli and generates sensor signals, which first pass through the data acquisition module and then the data processing module to complete basic filtering, and are then transmitted to the central decision-making system; the central decision-making system completes multi-sensor fusion and uses deep reinforcement learning to generate control strategies, and then the control signals are handed over to the control and execution unit to complete the movement of the robotic arm.

[0058] The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any technician in the relevant technical field, without departing from the scope of the technical solution of the present invention, makes any form of equivalent replacement or modification to the technical solution and technical content disclosed in the present invention, which does not depart from the content of the technical solution of the present invention and still falls within the protection scope of the present invention.

Claims

1. A Tai Chi Push Hands robotic arm based on electronic skin, characterized in that: It includes a robot body, an electronic skin sensing system wrapped in the robot body, a central decision system connected to the electronic skin sensing system, and a control execution system connected to the central decision system signal; The electronic skin sensing system is modularly designed and includes multiple sensing modules, each of which includes a front-end sensor element, a data processing unit and a data transmission unit. The data processing unit receives and processes the information of the front-end sensor element and transmits the information to the central decision-making system through the data transmission unit. The central decision system includes a contact pattern extraction unit and a decision unit. The contact pattern extraction unit receives the signal transmitted by the data transmission unit based on a multi-sensor fusion algorithm and extracts the contact semantics. The decision unit judges human intentions based on deep reinforcement learning and generates a control strategy. The control execution system includes a control unit and an execution unit connected to each other, the control unit receives the control strategy, generates a corresponding control signal, and transmits the control signal to the execution unit, the execution unit includes a motor, and the motor drives the robot arm to complete the corresponding action through the control signal; The multi-sensor fusion algorithm includes the following steps: S1: Decouple multiple targets and associate sensor information of the same event, cluster and classify different target data, obtain the contact time of the left and right arms of the robot, and then associate the data packets of the corresponding areas. When clustering and classifying different data, discard the cluster classification with too few activated units of the data; S2: Data alignment, performing spatiotemporal matching based on the position and time information in the data packet, unifying the position and time information into the robot global coordinate system and the global unified time node; S3: The pressure magnitude and direction when the human hand contacts the robotic arm are obtained based on the values ​​of the three-dimensional pressure sensor, and the total force and torque magnitude are calculated based on the orientation of the values ​​in the global coordinate system of the robotic arm; S4: Kalman filtering is performed on the contact force and torque to predict the motion trajectory of the human arm; S5: The prediction of the human arm motion trajectory and the magnitude of the contact force are put into the trained deep neural network. The output of the network is the predefined contact pattern, and finally the corresponding contact pattern is obtained; In S5, the process of deep neural network training includes the following steps: S51: adjusting the robotic arm to the teaching mode or the follow-up control mode, collecting the raw data of all sensors when the robotic arm and the human perform normal Tai Chi push-hands activities, aligning the group data in time and space, and obtaining the raw data set; initializing and constructing a deep neural network, and setting the initial parameters of the deep neural network. The structure of the deep neural network uses a conventional network structure, the number of input layer neurons is set according to the selected sensor, and the number of output layer neurons is set according to the electronic skin state quantity; S52: performing preprocessing, feature extraction and standardization calculation on the measurement data of the sensor in the original data set to obtain a training data set; the standardization calculation includes performing Kalman filtering on the data to obtain a motion trajectory; S53: Use the training data set to train the deep neural network to obtain a trained deep neural network.

2. The electronic skin-based Tai Chi Push Hands robotic arm according to claim 1, characterized in that: The front-end sensing element is one or more of a pressure sensor, a proximity sensor, a vibration sensor, a temperature sensor, an acceleration sensor, a magnetic sensor, a stretch sensor, a gyroscope and a magnetometer.

3. The electronic skin-based Tai Chi Push Hands robotic arm according to claim 2, characterized in that: The modular electronic skin sensing system is cut or spliced ​​into a whole piece of electronic skin of any shape. The whole piece of electronic skin is covered on the surface of the robotic arm with different shapes and curvatures. The robotic arm recognizes and collects various physical quantities of the objects it contacts through the electronic skin, and processes the various physical quantities through the data processing unit.

4. The electronic skin-based Tai Chi Push Hands robotic arm according to claim 3, characterized in that: The data processing unit is an MCU, and the data processing unit has a built-in data processing algorithm. The data processing unit performs data processing and adaptive filtering and denoising through the data processing algorithm. The electronic skin sensing system switches the algorithm automatically according to actual needs.

5. The electronic skin-based Tai Chi Push Hands robotic arm according to claim 4, characterized in that: The data transmission unit is a wired transmitter or a wireless transmitter. When the data transmission unit is a wired transmitter, the wired transmitter is provided with a plurality of ports, and adjacent electronic skins and the electronic skin and the contact pattern extraction unit are connected to each other through the ports and lines. When the data transmission unit is a wireless transmitter, adjacent electronic skins and the electronic skin and the contact pattern extraction unit are connected via Bluetooth or WiFi signals.

6. The electronic skin-based Tai Chi Push Hands robotic arm according to claim 5, characterized in that: The decision-making unit uses a deep reinforcement learning method based on value learning to train a deep Q network to obtain an optimal action value function, including the following steps: SA1: Set a reward function, which is related to the contact pattern and contact pressure, and the degree of consistency between the contact pressure and contact pattern and the preset trajectory and pattern is proportional to the score; SA2: Training deep Q-network using temporal difference algorithm; SA3: After training, the deep Q network outputs the optimal action for the next step, which is the output of the decision unit.

7. The electronic skin-based Tai Chi Push Hands robotic arm according to claim 6, characterized in that: The degree of fit is set with a threshold. When it is higher than the threshold, the contact pattern extracted by the contact pattern extraction unit is transmitted to the deep Q network in the decision unit; when it is lower than the threshold, the robotic arm issues a reminder and enters a compliant mode, in which the robotic arm moves completely according to the trajectory of the human arm.

8. The electronic skin-based Tai Chi Push Hands robotic arm according to claim 7, characterized in that: During the use of the Tai Chi Push Hands robotic arm, when the electronic skin senses excessive external force, the robotic arm stops moving and issues a reminder. After a delay of 5 seconds, the robotic arm automatically returns to the initial position of the Tai Chi Push Hands.

Citation Information

Patent Citations

  • Taiji push hand robot based on depth camera

    CN108621164A

  • Tai Ji pushing hand robot

    CN111496803A

  • Bionic flexible modular electronic skin system and control method thereof

    CN113447073A