CN112621714A - Upper limb exoskeleton robot control method and device based on LSTM neural network - Google Patents

Upper limb exoskeleton robot control method and device based on LSTM neural network Download PDF

Info

Publication number
CN112621714A
CN112621714A CN202011402755.6A CN202011402755A CN112621714A CN 112621714 A CN112621714 A CN 112621714A CN 202011402755 A CN202011402755 A CN 202011402755A CN 112621714 A CN112621714 A CN 112621714A
Authority
CN
China
Prior art keywords
exoskeleton robot
neural network
upper limb
human body
lstm neural
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011402755.6A
Other languages
Chinese (zh)
Inventor
李昀佶
王欣然
张震宇
芮岳峰
方略
王春雷
杨亚
范春辉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Micro Motor Research Institute 21st Research Institute Of China Electronics Technology Corp
Original Assignee
Shanghai Micro Motor Research Institute 21st Research Institute Of China Electronics Technology Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shanghai Micro Motor Research Institute 21st Research Institute Of China Electronics Technology Corp filed Critical Shanghai Micro Motor Research Institute 21st Research Institute Of China Electronics Technology Corp
Priority to CN202011402755.6A priority Critical patent/CN112621714A/en
Publication of CN112621714A publication Critical patent/CN112621714A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00—Programme-controlled manipulators
    • B25J9/0006—Exoskeletons, i.e. resembling a human figure
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J13/00—Controls for manipulators
    • B25J13/08—Controls for manipulators by means of sensing devices, e.g. viewing or touching devices
    • B25J13/085—Force or torque sensors
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J18/00—Arms
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00—Programme-controlled manipulators
    • B25J9/16—Programme controls
    • B25J9/1656—Programme controls characterised by programming, planning systems for manipulators
    • B25J9/1664—Programme controls characterised by programming, planning systems for manipulators characterised by motion, path, trajectory planning

Landscapes

  • Engineering & Computer Science (AREA)
  • Robotics (AREA)
  • Mechanical Engineering (AREA)
  • Human Computer Interaction (AREA)
  • Manipulator (AREA)

Abstract

The invention provides a control method and a device of an upper limb exoskeleton robot based on an LSTM neural network, wherein a robot arm of the upper limb exoskeleton robot moves along with the upper limb of a human body, and the method comprises the following steps: processing historical motion trail data of the human body through the constructed LSTM neural network prediction model, and predicting to obtain future motion state data of the human body; obtaining compensation torque required for controlling the upper limb exoskeleton robot according to the future motion state data of the human body and a pre-established dynamic model of the upper limb exoskeleton robot; acquiring the human-computer interaction force of the upper limb exoskeleton robot; and processing the compensation torque and the human-computer interaction force, and controlling the upper limb exoskeleton robot according to a processing result. The invention can realize the following of the upper limb exoskeleton robot to the human body movement under the condition of zero man-machine interaction force.

Description

Upper limb exoskeleton robot control method and device based on LSTM neural network
Technical Field
The invention relates to the technical field of exoskeleton robot control, in particular to an upper limb exoskeleton robot control method and device based on an LSTM neural network.
Background
At present, the main technical directions for upper limb exoskeleton robot research at home and abroad are as follows: a study focused on rehabilitation training to help patients with upper limb paralysis or muscle damage to recover function; the other type of the robot mainly aims at lifting the carrying direction, the express logistics field which is developed rapidly becomes an important civil market for application, the shell carrying and loading of armored artillery soldiers in the plateau oxygen-deficient environment becomes an important military market for application, and the problems that a courier and soldiers are labor-consuming in free-hand carrying, spinal damage is easy to occur and the like can be solved.
Most of existing upper limb exoskeleton robot control algorithms are force position control algorithms, Sensitivity Amplification Control (SAC) and admittance control. The force position control algorithm has high response speed, can detect the interaction force of people and compensate the interaction force in a targeted manner, but has high requirements on modeling precision. The Sensitivity Amplification Control (SAC) method follows the user's motion by directly compensating the dynamics of the moving parts, and is essentially an open-loop control method that cannot adapt to a changing load and is easily disturbed due to the lack of feedback on the state information of the human body in the system. The admittance control disclosed in patent application No. 201910188958.0 entitled "cooperative control method for upper limb exoskeleton with dynamic load compensation" is a method for adjusting force by speed control, and the algorithm thereof includes a plurality of feedback closed loops, belongs to cascade control, but the boosting effect of the speed control is not significant enough. The current control algorithm can not realize real-time following of human body movement, and has certain lag in control feedback.
Therefore, there is a need to provide an improved solution to the above-mentioned deficiencies in the prior art.
Disclosure of Invention
The invention aims to provide a control method and a control device of an upper limb exoskeleton robot based on an LSTM neural network, so as to solve the existing technical problems.
In order to solve the technical problems, the invention provides the following technical scheme:
an upper limb exoskeleton robot control method based on an LSTM neural network, wherein a robot arm of the upper limb exoskeleton robot follows the motion of an upper limb of a human body, and the method comprises the following steps:
processing historical motion trail data of the human body through the constructed LSTM neural network prediction model, and predicting to obtain future motion state data of the human body;
obtaining compensation torque required for controlling the upper limb exoskeleton robot according to the future motion state data of the human body and a pre-established dynamic model of the upper limb exoskeleton robot;
acquiring the human-computer interaction force of the upper limb exoskeleton robot;
and processing the compensation torque and the human-computer interaction force, and controlling the upper limb exoskeleton robot according to a processing result.
Further, the step of predicting the historical motion trajectory data of the human body through the constructed LSTM neural network prediction model to obtain the future motion state data of the human body comprises the following steps:
measuring motion state data of the robot arm through an inertia measuring unit, taking the motion state data of the robot arm in a set time period as historical track data of a human body, and taking the motion state data of the robot arm at the latest moment as real-time motion state data of the human body;
constructing an LSTM neural network prediction model based on the historical track data of the human body;
and predicting to obtain the future motion state data of the human body according to the real-time motion state data of the human body and the LSTM neural network prediction model.
Further, the motion state data includes an articulation angle, an articulation angular velocity, and an articulation angular acceleration.
Further, training the historical trajectory data of the human body by a Bayesian normalization method to obtain an LSTM neural network prediction model.
Further, the compensation moment comprises inertia moment, gravity moment and friction moment.
Further, according to the future motion state data of the human body and a pre-established dynamic model of the upper limb exoskeleton robot, the gravitational moment required for controlling the upper limb exoskeleton robot is obtained through an inverse dynamic method.
Further, the friction torque and the inertia torque required for controlling the upper limb exoskeleton robot are obtained through a parameter identification method according to the future motion state data of the human body and the dynamic model of the upper limb exoskeleton robot.
Further, the processing the compensation torque and the human-computer interaction force, and controlling the upper limb exoskeleton robot according to the processing result includes: processing the compensation torque and the human-computer interaction force according to a torque balance equation of the upper limb exoskeleton robot to obtain a motor torque; and controlling the upper limb exoskeleton robot according to the motor torque.
Further, the compensation torque and the human-computer interaction force are processed according to a torque balance equation of the upper limb exoskeleton robot to obtain a motor torque; and the control of the upper limb exoskeleton robot according to the motor torque comprises the following steps:
according to a moment balance equation of the upper limb exoskeleton robot, firstly assuming that the human-computer interaction force is zero, and then calculating with the compensation moment to obtain a processing result, wherein the processing result is motor moment;
and controlling the upper limb exoskeleton robot through the motor torque according to the predicted joint movement angular speed as a target.
In order to solve the above technical problems, the present invention further provides:
an upper limb exoskeleton robot control device based on an LSTM neural network comprises a processor, a memory and a computer program stored in the memory and capable of running on the processor, wherein the processor realizes the steps of the upper limb exoskeleton robot control method based on the LSTM neural network when executing the computer program.
Compared with the closest prior art, the technical scheme provided by the invention has the following beneficial effects:
according to the upper limb exoskeleton robot control method based on the LSTM neural network, historical motion track data of a human body is predicted through a constructed LSTM neural network prediction model, and future motion state data of the human body is obtained; obtaining compensation torque required by controlling the upper limb exoskeleton robot according to the future motion state data of the human body and the dynamic model of the upper limb exoskeleton robot; and processing the compensation torque and the acquired human-computer interaction force of the upper limb exoskeleton robot, and controlling the upper limb exoskeleton robot according to a processing result. The control method can realize the following of the upper limb exoskeleton robot to the human body movement under the condition that the human-computer interaction force is zero, and is simple and better in accuracy.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the invention and, together with the description, serve to explain the invention and not to limit the invention. Wherein:
FIG. 1 is a flow chart of a control method of an upper limb exoskeleton robot based on an LSTM neural network according to the invention;
FIG. 2 is a flow chart of the present invention for constructing an LSTM neural network prediction model and controlling through the LSTM neural network model;
FIG. 3 is a flow chart of the present invention for obtaining future motion state data of a human body through an LSTM neural network prediction model;
FIG. 4 is a topological structure diagram of the LSTM neural network of the present invention;
fig. 5 is a control schematic diagram of the upper limb exoskeleton robot based on the LSTM neural network according to the present invention.
Detailed Description
The present invention will be described in detail below with reference to the embodiments with reference to the attached drawings. The various examples are provided by way of explanation of the invention, and not limitation of the invention. In fact, it will be apparent to those skilled in the art that modifications and variations can be made in the present invention without departing from the scope or spirit thereof. For instance, features illustrated or described as part of one embodiment, can be used with another embodiment to yield a still further embodiment. It is therefore intended that the present invention encompass such modifications and variations as fall within the scope of the appended claims and equivalents thereof.
In this application, upper limbs ectoskeleton robot includes body and two robotic arms, and two robotic arms are connected to body upper portion both sides swing joint respectively. In the application process of the upper limb exoskeleton robot, the upper limb exoskeleton robot bears fixed load in a bearing mode, namely the two robot arms bear the fixed load; the two robot arms are close to or the same as the moving track of the human body, namely the robot arms follow the upper limbs of the human body to move, the control aim of the upper limb exoskeleton robot based on the LSTM neural network is to realize that the two robot arms and the human body move completely, and the complete following means that the moving track of the robot arms and the moving track of the human body are completely the same, namely the acting force of the robot arms and the human body is zero, or the human-computer interaction force of the robot arms is zero).
It should be noted that, since the present application mainly controls the robot arm of the upper limb exoskeleton robot, the human-computer interaction force of the upper limb exoskeleton robot described in the present application refers to the acting force between the robot arm of the upper limb exoskeleton robot and the upper limb of the human body; meanwhile, the dynamic model of the upper limb exoskeleton robot is specifically a dynamic model of a robot arm of the upper limb exoskeleton robot.
Fig. 1 and fig. 2 show a flowchart of a control method of an upper limb exoskeleton robot based on an LSTM neural network according to the present invention, and the specific control method includes:
step 1, processing historical motion trail data of a human body through a constructed LSTM neural network prediction model, and predicting to obtain future motion state data of the human body;
the method comprises the steps that two IMUs (inertial measurement units) are respectively installed on two robot arms of an upper-limb exoskeleton robot, motion state data of the robot arms can be measured through the IMUs, the measured motion state data of the robot arms serve as historical motion trajectory data of a human body, the future motion state data of the human body can be predicted through an LSTM neural network prediction algorithm based on the historical motion trajectory data of the human body, and in the embodiment of the application, the predicted motion state data of the human body is 0.5-1 second in the future.
In this step, the process of obtaining the future motion state data of the human body through the LSTM neural network prediction algorithm based on the historical motion trajectory data of the human body is shown in fig. 3 as follows:
step 11, obtaining motion state data of the robot arm through measurement of an Inertial Measurement Unit (IMU), wherein the motion state data of the robot arm in a set time period is used as historical track data of a human body, and the motion state data of the robot arm at the latest moment is used as real-time motion state data of the human body;
an Inertial Measurement Unit (IMU) is a device that measures the three-axis attitude angles (or angular rates) and acceleration of an object. The IMU comprises three single-axis accelerometers and three single-axis gyroscopes, the accelerometers detect acceleration signals of an object on three independent axes of a carrier coordinate system, the gyroscopes detect angular velocity signals of a carrier relative to a navigation coordinate system, and angular velocity and acceleration of the object in a three-dimensional space.
In an embodiment of the application, the motion state data comprises an articulation angle, an articulation angular acceleration and an articulation angular velocity.
Step 12, constructing and obtaining an LSTM neural network prediction model based on the historical track data of the human body;
and processing and dividing the historical track data of the human body to obtain a training set and a test set, wherein the training set is used for training the LTSM neural network, and the test set is used for verifying data and the LSTM neural network.
In the process of training and constructing an LSTM neural network prediction model, historical motion track data (such as joint motion angle, joint motion angular velocity and joint motion angular acceleration) of a human body are obtained from an Inertial Measurement Unit (IMU) and used as training data, a set of the training data is divided into a training set and a testing set, 70% of the training set is used as the training data, and weight values and threshold values of neural connections of each layer of a gateway in the LSTM neural network are adjusted; of which 15% was used as the first test set as the validation data for preventing the overfitting phenomenon; and taking 15% of the test data as a second test set to test and verify the prediction precision of the trained LSTM neural network prediction model.
It should be noted that, the process of training data and adjusting the training thereof by using the LSTM neural network is prior art, and this process is not described in detail in the embodiments of the present application, and only the LSTM neural network prediction model of the present application is described below.
The LSTM neural network prediction model is shown by the following equation:
y(t+1)＝f(y(t)，y(t-1)，...y(t-dy+1))
wherein f is a non-linear function, dyIs the delay order, y (t) is the historical motion trail data of the human body at the time t, y (t-1) is the historical motion trail data of the human body at the time t-1, y (t-d)y+1) is t-dyThe motion trajectory data of the human body at the moment +1, and y (t +1) is the future motion state data of the human body at the moment t + 1.
FIG. 4 is a diagram of the topology of the LSTM neural network of the present application. The LSTM neural network is additionally provided with an input gate, a forgetting gate and an output gate which selectively retain required information at each moment, and simultaneously, the unit state c is added for storing the current hidden state, so that the previous information can be continuously transmitted backwards without disappearance along with the continuous superposition of the input sequence of the hidden layer under a new time state.
For the prediction model of the LSTM neural network, the left sides y (t-n) -y (t-1) are shoulder joint angle data of the previous n steps, the middle part is a hidden layer, and the output of each neuron needs to be calculated according to the following formula.
Figure BDA0002812988280000061
In the function L (·), the hidden states include an input gate i, a forgetting gate f, an output gate o, and a cell state c;
determined by the following equation:
an input gate:
Figure BDA0002812988280000062
forget the door:
Figure BDA0002812988280000063
Figure BDA0002812988280000064
an output gate:
Figure BDA0002812988280000065
long memory:
Figure BDA0002812988280000066
short memory: h ist＝Ot*tanh(Ct)
In the formula, Wi，Wc，WoThe weights of the input gate, cell state, output gate, respectively; h isi tIs an implicit state; bi，bc，boThe input gate, the cell state, and the output gate bias, respectively, and C — is abstracted information extracted during the calculation.
In the embodiment of the application, a Long Short-Term Memory Network (LSTM) prediction algorithm is memorized, so that the LSTM prediction algorithm has the characteristics of outputting feedback and processing a time sequence prediction problem, and can also selectively memorize information to be memorized for a Long time through a unit state c structure in the LSTM prediction algorithm, so that the prediction precision can be improved. The LSTM neural network is additionally provided with an input gate, a forgetting gate and an output gate to selectively retain required information at each moment, and meanwhile, the unit state c is added to store the current hidden state, so that the situation that the previous information can be continuously propagated backwards without disappearing as the hidden layer is continuously overlapped with an input sequence in a new time state is ensured.
The number of the hidden layer neurons affects the performance of the network, the number is small, the fitting effect is poor, the number is large, weight transition fitting is easy to occur, and the generalization performance is poor. In the embodiment of the present application, the number of neurons in the hidden layer is determined by the following formula.
Figure BDA0002812988280000071
In the formula, NhidIs a hidden layer neuron number estimated value; n is a radical ofinAnd NoutThe numbers of neurons in an input layer and an output layer are respectively; m is a training sample; a is a constant and takes a value of 5-10; r is a constant and takes a value of 1-10.
In the embodiment of the application, the historical trajectory data of the human body is trained through a Bayesian normalization method to obtain an LSTM neural network prediction model. The Bayesian normalization method is selected to serve as a training algorithm of the neural network, the generalization ability of the LSTM neural network prediction model can be improved by training the training data, namely, the generalization ability of the LSTM neural network prediction model can be effectively improved by training the LSTM neural network prediction model by the training data through the Bayesian normalization method.
In the embodiment of the application, the prediction accuracy of the LSTM neural network prediction model is evaluated by Mean Square Error (MSE), wherein the MSE represents the mean square error between a real value and a predicted value, and the smaller the MSE value is, the higher the prediction accuracy is. And when the MSE value is smaller than the set error target value delta, finishing the training of the LSTM neural network prediction model, if the MSE value is larger than the set error target value delta, readjusting the network parameters of the LSTM neural network prediction model, and continuing to train the LSTM neural network prediction model until the MSE value is smaller than the set error target value delta, and finishing the training of the LSTM neural network prediction model.
In training, the LSTM neural network can continuously adjust the weight and threshold parameters among neurons according to data and corresponding labels, establish the mapping relation between input data and output labels, achieve self-learning self-adaptation, have high nonlinear characteristics and good robustness and fault tolerance, and play a role in predicting human body carrying intentions, thereby playing a role in more optimally controlling the upper limb exoskeleton.
Step 13, predicting to obtain future motion state data of the human body according to the real-time motion state data of the human body and the LSTM neural network prediction model;
in the embodiment of the application, the real-time motion state data of the human body is input into the LSTM neural network prediction model, so that the future motion state data of the human body can be obtained in a prediction mode.
When single step prediction is performed, the input of the prediction model is dyThe historical motion trajectory data (information on the joint angle of the shoulder joint) are represented as y (t), y (t-1), … y (t-d)y+1), the human motion state (joint motion) of the next second can be predicted by the LSTM neural network prediction modelAngle) outputs y (t +1), and the input values are real values at this time.
When multi-step prediction is carried out, the output prediction value can be fed back to be input of the next step, and the output of the next step is predicted. If y (t +2) is predicted at the t second, the input of the prediction model is a prediction angle y (t +1) and historical angle data y (t), y (t-1),. y (t-d)y+2) and is likewise dyInput data, but including 1 predictor and d y1 true value. And (4) through the prediction formula, multi-step prediction can be realized through repeated iteration.
In the embodiment of the application, the human motion trajectory prediction means that the motion state data of the next time period is predicted according to the historical trajectory data of the human, and is mainly expressed on the angle change of the joint. Therefore, the input data of the LSTM neural network prediction model is the joint motion angle information in the historical trajectory data, and the output data of the LSTM neural network prediction model is the joint motion angle information obtained by prediction. As another embodiment, the joint motion angular acceleration, joint motion angular velocity, and the like may be predicted.
In the application, the LSTM neural network has good nonlinear fitting capability and capability of processing long-term time sequences, and has great advantages in processing nonlinear and time-varying problems, so that data with high nonlinear and time-varying characteristics of human motion (robot arms) are processed, and future motion state data of the human body can be accurately predicted.
Step 2, obtaining compensation torque required for controlling the upper limb exoskeleton robot according to the future motion state data of the human body and a pre-established dynamic model of the upper limb exoskeleton robot; the compensation moment comprises inertia moment, gravity moment and friction moment;
in the embodiment of the present application, a pre-established dynamic model of the upper extremity exoskeleton robot is established according to a specific structural configuration of the robot, and a newton-eulerian method and a lagrangian method are adopted to establish the dynamic model of the upper extremity exoskeleton robot, where the established specific method is the prior art and is not described in detail in the embodiment of the present application.
Substituting the future motion state data (angle, angular velocity and angular acceleration) of the human body into a dynamic model of the upper limb exoskeleton robot, and analyzing and solving the dynamic model to obtain compensation torque required by controlling the upper limb exoskeleton robot; the compensation moment comprises inertia moment, gravity moment and friction moment.
Obtaining a constraint equation of the upper limb exoskeleton robot according to the dynamic model, and then solving forward and inverse kinematics, wherein in the embodiment of the application, the dynamic model is solved through inverse kinematics to obtain the compensation torque required by controlling the upper limb exoskeleton robot; or solving the dynamic model through a parameter identification method to obtain the compensation torque required by controlling the upper limb exoskeleton robot.
In the embodiment of the application, the gravity moment tau to be compensated is obtained through inverse dynamic calculationGravity force(ii) a The friction torque tau to be compensated can be obtained by a parameter identification methodFrictional forceAnd moment of inertia τForce of inertia。
Step 3, acquiring the human-computer interaction force of the upper limb exoskeleton robot;
in the embodiment of the application, the human-computer interaction force of the upper limb exoskeleton robot is detected through the pressure strain sensor, and comprises the human-computer interaction force of the first robot arm and the human-computer interaction force of the second robot arm.
The four pressure strain sensors are respectively arranged at the upper side and the lower side of the two robot arms of the upper-limb exoskeleton robot, and for convenience of the following description, the robot arm at the left side is defined as a first robot arm, and the robot arm at the right side is defined as a second robot arm.
Specifically, a first pressure strain sensor is arranged on the first robot arm, a second pressure strain sensor is arranged under the first robot arm, a third pressure strain sensor is arranged on the second robot arm, and a fourth pressure strain sensor is arranged under the second robot arm. Correspondingly, the pressure detected by the first pressure strain sensor is a first pressure, the pressure detected by the second pressure strain sensor is a second pressure, the pressure detected by the third pressure strain sensor is a third pressure, and the pressure detected by the fourth pressure strain sensor is a fourth pressure.
For the first machine arm, determining the man-machine interaction force of the first machine arm through the first pressure and the second pressure, and when the first pressure is zero and the second pressure is not zero, taking the second pressure value as the man-machine interaction force of the first machine arm; when the first pressure is not zero and the second pressure is zero, taking the first pressure value as the man-machine interaction force of the first robot arm; and when the first pressure and the second pressure are both zero, the man-machine interaction force of the first machine arm is zero.
Similarly, for the second robot arm, determining the human-computer interaction force of the first robot arm through the third pressure and the fourth pressure, and when the third pressure is zero and the fourth pressure is not zero, taking the third pressure value as the human-computer interaction force of the second robot arm; when the third pressure is not zero and the fourth pressure is zero, taking the third pressure value as the human-computer interaction force of the second robot arm; and when the third pressure and the fourth pressure are both zero, the human-computer interaction force of the second robot arm is zero.
When the pressure strain sensor on the upper surface of the robot arm senses the pressure (the first pressure or the third pressure), the output of the motor becomes larger in a qualitative sense, and when the pressure strain sensor on the lower surface of the robot arm senses the pressure (the second pressure or the fourth pressure), the output of the motor becomes smaller.
The load of the human body is directly loaded on the upper-limb exoskeleton robot, but not directly contacted with the human body, so that when the human body and the robot arm are in a complete following state, the human-computer interaction force is zero, and the control target of the robot arm is that the human-computer interaction force of the first robot arm and the human-computer interaction force of the second robot arm are both zero; when the robot arm does not completely follow the human motion, the human-computer interaction force of the robot arm is not zero, so that in the process of controlling the robot arm, when the human-computer interaction force of the first robot arm or the human-computer interaction force of the second robot arm is not zero, the motor is required to be controlled to continuously adjust the output torque until the pressure value detected by the corresponding pressure strain sensor is zero, and at the moment, the adjustment target is reached, namely the upper limb exoskeleton robot completely follows the motion track of the human body.
In the embodiment of the application, the human-computer interaction force of the robot arm is zero, which is an optimal and ideal state, but considering the influence of actual feedback regulation control, the human-computer interaction force of the robot arm can be a value satisfying a certain range (smaller than a set value close to zero), and at the moment, the exoskeleton upper limb robot generates an assistance effect on a human body when the human body bears a weight and moves along with the movement consciousness of the human body.
And 4, processing the compensation torque and the human-computer interaction force, and controlling the upper limb exoskeleton robot according to a processing result.
In the embodiment of the application, the compensation moment and the human-computer interaction force are processed according to a moment balance equation of the upper limb exoskeleton robot, and the upper limb exoskeleton robot is controlled according to a processing result. The specific process comprises the following steps:
step 41, according to a moment balance equation of the upper limb exoskeleton robot, firstly assuming that the human-computer interaction force is zero, and then calculating with the compensation moment to obtain a processing result, wherein the processing result is a motor moment;
in the embodiment of the present application, the moment balance equation is
τElectric machine+τInteraction＝τForce of inertia+τGravity force+τFrictional force
When the upper limb exoskeleton robot is to move completely along with the movement of the human body, tau is required to be satisfiedInteractionTherefore, assuming that the human-computer interaction force is zero, the magnitude of the motor output torque, that is, the motor torque, is calculated.
Step 42, controlling the upper limb exoskeleton robot through the motor torque according to the predicted joint movement angular speed as a target;
the predicted joint movement angular speed is used as the expected speed which is expected to be achieved by the upper limb exoskeleton robot control, and the joint movement angular speed expression is
Figure BDA0002812988280000111
In the formula (I), the compound is shown in the specification,
Figure BDA0002812988280000112
f is the human-computer interaction force and J is the matrix constant for the predicted joint motion angular velocity; d is the damping coefficient of the damping material,
Figure BDA0002812988280000113
is the actual articulation angular velocity.
When the actual angular velocity reaches the predicted angular velocity, the magnitude of the human-computer interaction force is just zero.
In the embodiment of the application, the controller controls the upper limb exoskeleton robot according to the motor torque. The controller is a PD controller. Based on the calculated magnitudes of the respective moments to be compensated, the use of the PD controller can improve the rapidity of the system adjustment.
Figure BDA0002812988280000114
In the formula, τInteractionFor human-computer interaction, KPIs a coefficient of proportionality that is,
Figure BDA0002812988280000115
in order to predict the angular velocity of the joint motion,
Figure BDA0002812988280000116
to actual angular velocity of joint motion, KDIs a differential coefficient;
Figure BDA0002812988280000117
actual joint angular acceleration;
Figure BDA0002812988280000118
is the predicted angular acceleration of the joint motion.
At the moment, the motion state of the joint is directly influenced by the man-machine interaction force, and the motion trend of the human body can be directly reflected on the motion of the joint, so that the control purpose can be well achieved through the control method.
The embodiment of the device is as follows:
the upper limb exoskeleton robot control device based on the LSTM neural network comprises a processor, a memory and a computer program stored in the memory and capable of running on the processor, wherein the processor realizes the steps of the upper limb exoskeleton robot control method based on the LSTM neural network when executing the computer program.
The control schematic diagram of the control device is shown in fig. 5, and the pressures of the two robot arms are detected by different pressure strain sensors respectively, so that the man-machine interaction force is obtained; the motion state data are detected through the IMU, then the PD controller processes the human-computer interaction force and the motion state data, and the motors of the two robot arms are controlled through the drivers according to the processing result, so that the control of the upper limb exoskeleton robot is realized. The steps of the control method for the upper limb exoskeleton robot based on the LSTM neural network are introduced in the embodiment of the method, and are not described in detail herein.
In summary, the present invention can achieve the following beneficial technical effects:
in the control process, the data of the angular velocity change of the movement of the shoulder joint of the human body are obtained in real time based on the LSTM neural network prediction model obtained through training, so that the movement state of the human body in the future within a short time is predicted, and the control is performed according to the prediction result, so that the joint movement of the upper limb exoskeleton robot can completely follow the movement of the human body.
The method is based on the LSTM neural network prediction model, more accurate future motion state data can be obtained through prediction, therefore, in the control process of the method, the motor torque size obtained is controlled in a mode that the size of the human-computer interaction force is zero, the control process is simple, fluctuation in the control process can be reduced, and the motion stability in the control process of the upper limb exoskeleton robot is improved.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. An upper limb exoskeleton robot control method based on an LSTM neural network, wherein a robot arm of the upper limb exoskeleton robot follows the motion of an upper limb of a human body, and the method comprises the following steps:
processing historical motion trail data of the human body through the constructed LSTM neural network prediction model, and predicting to obtain future motion state data of the human body;
obtaining compensation torque required for controlling the upper limb exoskeleton robot according to the future motion state data of the human body and a pre-established dynamic model of the upper limb exoskeleton robot;
acquiring the human-computer interaction force of the upper limb exoskeleton robot;
and processing the compensation torque and the human-computer interaction force, and controlling the upper limb exoskeleton robot according to a processing result.
2. The LSTM neural network-based upper extremity exoskeleton robot control method of claim 1, wherein predicting historical motion trajectory data of the human body by using the constructed LSTM neural network prediction model to obtain future motion state data of the human body comprises:
measuring motion state data of the robot arm through an inertia measuring unit, taking the motion state data of the robot arm in a set time period as historical track data of a human body, and taking the motion state data of the robot arm at the latest moment as real-time motion state data of the human body;
constructing an LSTM neural network prediction model based on the historical track data of the human body;
and predicting to obtain the future motion state data of the human body according to the real-time motion state data of the human body and the LSTM neural network prediction model.
3. The LSTM neural network-based upper extremity exoskeleton robot control method of claim 1 or 2, wherein the kinematic state data includes joint motion angle, joint motion angular velocity and joint motion angular acceleration.
4. The LSTM neural network-based upper extremity exoskeleton robot control method of claim 2, wherein the LSTM neural network prediction model is obtained by training the historical trajectory data of the human body through a bayesian normalization method.
5. The LSTM neural network-based upper extremity exoskeleton robot control method of claim 1, wherein the compensation moments include moments of inertia, moments of gravity and moments of friction.
6. The LSTM neural network-based upper extremity exoskeleton robot control method of claim 5, wherein the gravitational moment required to control the upper extremity exoskeleton robot is obtained by inverse dynamics method according to the future motion state data of the human body and the pre-established dynamics model of the upper extremity exoskeleton robot.
7. The LSTM neural network based upper extremity exoskeleton robot control method of claim 5, wherein the obtained friction torque and inertia torque required for controlling the upper extremity exoskeleton robot are obtained through a parameter identification method according to the future motion state data of the human body and the dynamic model of the upper extremity exoskeleton robot.
8. The LSTM neural network-based upper extremity exoskeleton robot control method of claim 1, wherein said processing the compensation torque and the human-computer interaction force, and controlling the upper extremity exoskeleton robot according to the processing result comprises: processing the compensation torque and the human-computer interaction force according to a torque balance equation of the upper limb exoskeleton robot to obtain a motor torque; and controlling the upper limb exoskeleton robot according to the motor torque.
9. The LSTM neural network-based upper extremity exoskeleton robot control method of claim 8, wherein the compensation torque and the human-computer interaction force are processed according to a torque balance equation of the upper extremity exoskeleton robot to obtain a motor torque; and the control of the upper limb exoskeleton robot according to the motor torque comprises the following steps:
according to a moment balance equation of the upper limb exoskeleton robot, firstly assuming that the human-computer interaction force is zero, and then calculating with the compensation moment to obtain a processing result, wherein the processing result is motor moment;
and controlling the upper limb exoskeleton robot through the motor torque according to the predicted joint movement angular speed as a target.
10. An upper extremity exoskeleton robot control device based on an LSTM neural network, comprising a processor, a memory and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the upper extremity exoskeleton robot control method based on an LSTM neural network according to any one of claims 1 to 9 when executing the computer program.
CN202011402755.6A 2020-12-02 2020-12-02 Upper limb exoskeleton robot control method and device based on LSTM neural network Pending CN112621714A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011402755.6A CN112621714A (en) 2020-12-02 2020-12-02 Upper limb exoskeleton robot control method and device based on LSTM neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011402755.6A CN112621714A (en) 2020-12-02 2020-12-02 Upper limb exoskeleton robot control method and device based on LSTM neural network

Publications (1)

Publication Number Publication Date
CN112621714A true CN112621714A (en) 2021-04-09

Family

ID=75308091

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011402755.6A Pending CN112621714A (en) 2020-12-02 2020-12-02 Upper limb exoskeleton robot control method and device based on LSTM neural network

Country Status (1)

Country Link
CN (1) CN112621714A (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113084813A (en) * 2021-04-13 2021-07-09 中国科学院自动化研究所 Robot motion control method for optimally constructing constraint force field based on muscle parameters
CN113197752A (en) * 2021-04-30 2021-08-03 华中科技大学 Limb gravity dynamic compensation method of upper limb rehabilitation robot
CN113478486A (en) * 2021-07-12 2021-10-08 上海微电机研究所(中国电子科技集团公司第二十一研究所) Robot motion parameter self-adaptive control method and system based on deep reinforcement learning
CN113547524A (en) * 2021-08-16 2021-10-26 长春工业大学 Man-machine interaction control method of upper limb exoskeleton robot
CN114406983A (en) * 2021-12-06 2022-04-29 中国科学院深圳先进技术研究院 Adaptive admittance control method and related device for lower limb exoskeleton robot
CN115755592A (en) * 2023-01-10 2023-03-07 常熟理工学院 Multi-mode control method for adjusting motion state of three-degree-of-freedom exoskeleton and exoskeleton

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101336848A (en) * 2008-08-22 2009-01-07 中国人民解放军海军航空工程学院 Man machine exoskeleton system and force control device thereof
CN102202613A (en) * 2008-09-04 2011-09-28 Iwalk股份有限公司 Hybrid terrain-adaptive lower-extremity systems
CN103171614A (en) * 2011-12-23 2013-06-26 联创汽车电子有限公司 Torque control device used in electric power steering system
US9149938B1 (en) * 2014-04-11 2015-10-06 Harris Corporation Robotic exoskeleton with adaptive viscous user coupling
CN107397649A (en) * 2017-08-10 2017-11-28 燕山大学 A kind of upper limbs exoskeleton rehabilitation robot control method based on radial base neural net
CN110125909A (en) * 2019-05-22 2019-08-16 南京师范大学镇江创新发展研究院 A kind of multi-information fusion human body exoskeleton robot Control protection system
CN111590544A (en) * 2020-04-10 2020-08-28 南方科技大学 Method and device for determining output force of exoskeleton

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101336848A (en) * 2008-08-22 2009-01-07 中国人民解放军海军航空工程学院 Man machine exoskeleton system and force control device thereof
CN102202613A (en) * 2008-09-04 2011-09-28 Iwalk股份有限公司 Hybrid terrain-adaptive lower-extremity systems
CN103171614A (en) * 2011-12-23 2013-06-26 联创汽车电子有限公司 Torque control device used in electric power steering system
US9149938B1 (en) * 2014-04-11 2015-10-06 Harris Corporation Robotic exoskeleton with adaptive viscous user coupling
CN107397649A (en) * 2017-08-10 2017-11-28 燕山大学 A kind of upper limbs exoskeleton rehabilitation robot control method based on radial base neural net
CN110125909A (en) * 2019-05-22 2019-08-16 南京师范大学镇江创新发展研究院 A kind of multi-information fusion human body exoskeleton robot Control protection system
CN111590544A (en) * 2020-04-10 2020-08-28 南方科技大学 Method and device for determining output force of exoskeleton

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
张国帅: "基于深度学习的多目标跟踪关联模型设计", 《中国硕士学位论文全文数据库（电子期刊）信息科技辑》 *
李洪武: "下肢助力外骨骼人机交互力传感器设计及控制方法研究", 《中国硕士学位论文全文数据库（电子期刊）信息科技辑》 *
邵建军: "基于神经网络和遗传算法的激光选区熔化成形工艺优化研究", 《中国硕士学位论文全文数据库（电子期刊）基础科学辑》 *
龙亿: "下肢外骨骼人体运动预测与人机协调控制技术研究", 《中国博士学位论文全文数据库（电子期刊）信息科技辑》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113084813A (en) * 2021-04-13 2021-07-09 中国科学院自动化研究所 Robot motion control method for optimally constructing constraint force field based on muscle parameters
CN113084813B (en) * 2021-04-13 2022-05-03 中国科学院自动化研究所 Robot motion control method for optimally constructing constraint force field based on muscle parameters
CN113197752A (en) * 2021-04-30 2021-08-03 华中科技大学 Limb gravity dynamic compensation method of upper limb rehabilitation robot
CN113478486A (en) * 2021-07-12 2021-10-08 上海微电机研究所(中国电子科技集团公司第二十一研究所) Robot motion parameter self-adaptive control method and system based on deep reinforcement learning
CN113478486B (en) * 2021-07-12 2022-05-17 上海微电机研究所(中国电子科技集团公司第二十一研究所) Robot motion parameter self-adaptive control method and system based on deep reinforcement learning
CN113547524A (en) * 2021-08-16 2021-10-26 长春工业大学 Man-machine interaction control method of upper limb exoskeleton robot
CN113547524B (en) * 2021-08-16 2022-04-22 长春工业大学 Man-machine interaction control method of upper limb exoskeleton robot
CN114406983A (en) * 2021-12-06 2022-04-29 中国科学院深圳先进技术研究院 Adaptive admittance control method and related device for lower limb exoskeleton robot
CN115755592A (en) * 2023-01-10 2023-03-07 常熟理工学院 Multi-mode control method for adjusting motion state of three-degree-of-freedom exoskeleton and exoskeleton

Similar Documents

Publication Publication Date Title
CN112621714A (en) Upper limb exoskeleton robot control method and device based on LSTM neural network
CN111281743B (en) Self-adaptive flexible control method for exoskeleton robot for upper limb rehabilitation
CN111546315B (en) Robot flexible teaching and reproducing method based on human-computer cooperation
US8504208B2 (en) Mobile object controller and floor surface estimator
Salaris et al. Online optimal perception-aware trajectory generation
Kuo Trajectory and heading tracking of a mecanum wheeled robot using fuzzy logic control
Peng et al. Robust adaptive motion/force control scheme for crawler-type mobile manipulator with nonholonomic constraint based on sliding mode control approach
CN113377121A (en) Aircraft intelligent disturbance rejection control method based on deep reinforcement learning
Li et al. An online trajectory planning method of a flexible-link manipulator aiming at vibration suppression
Lin et al. Adaptive critic anti-slip control of wheeled autonomous robot
CN114310851B (en) Dragging teaching method of robot moment-free sensor
Xu et al. Whole-body stability control with high contact redundancy for wheel-legged hexapod robot driving over rough terrain
Tan et al. A cerebellum-inspired network model and learning approaches for solving kinematic tracking control of redundant manipulators
Chertopolokhov et al. Averaged sub-gradient integral sliding mode control design for cueing end-effector acceleration of a two-link robotic arm
CN113370208A (en) Control method and system for reconfigurable mechanical arm subsystem
JP2012245574A (en) Moving body control device
Wang et al. Balance equilibrium manifold and control of rider-bikebot systems
Zadeh et al. LQR motion control and analysis of a prototype spherical robot
CN116638507A (en) Teleoperation control method combining self-adaptive impedance control and predictive control
WO2023020036A1 (en) Redundant manipulator tracking control method based on echo state network
CN116047888A (en) Control method of self-balancing vehicle based on BP neural network PID
CN113043283B (en) Robot tail end external force estimation method
CN113515044B (en) Learning prediction-based tracking control method and device for wheeled mobile robot
Zhang et al. Trajectory-tracking control of robotic systems via deep reinforcement learning
WO2019095108A1 (en) Robot imitation learning method and apparatus, robot and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20210409

RJ01 Rejection of invention patent application after publication