CN114028164A - Rehabilitation robot control method and device and rehabilitation robot - Google Patents

Rehabilitation robot control method and device and rehabilitation robot Download PDF

Info

Publication number
CN114028164A
CN114028164A CN202111371350.5A CN202111371350A CN114028164A CN 114028164 A CN114028164 A CN 114028164A CN 202111371350 A CN202111371350 A CN 202111371350A CN 114028164 A CN114028164 A CN 114028164A
Authority
CN
China
Prior art keywords
training
trained
neural network
network model
joint
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
CN202111371350.5A
Other languages
Chinese (zh)
Inventor
孙维
黄冠
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.)
Shenzhen Huaquejing Medical Technology Co ltd
Original Assignee
Shenzhen Huaquejing Medical Technology Co ltd
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 Shenzhen Huaquejing Medical Technology Co ltd filed Critical Shenzhen Huaquejing Medical Technology Co ltd
Priority to CN202111371350.5A priority Critical patent/CN114028164A/en
Publication of CN114028164A publication Critical patent/CN114028164A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61H—PHYSICAL THERAPY APPARATUS, e.g. DEVICES FOR LOCATING OR STIMULATING REFLEX POINTS IN THE BODY; ARTIFICIAL RESPIRATION; MASSAGE; BATHING DEVICES FOR SPECIAL THERAPEUTIC OR HYGIENIC PURPOSES OR SPECIFIC PARTS OF THE BODY
    • A61H1/00—Apparatus for passive exercising; Vibrating apparatus; Chiropractic devices, e.g. body impacting devices, external devices for briefly extending or aligning unbroken bones
    • A61H1/02—Stretching or bending or torsioning apparatus for exercising
    • A61H1/0274—Stretching or bending or torsioning apparatus for exercising for the upper limbs
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61H—PHYSICAL THERAPY APPARATUS, e.g. DEVICES FOR LOCATING OR STIMULATING REFLEX POINTS IN THE BODY; ARTIFICIAL RESPIRATION; MASSAGE; BATHING DEVICES FOR SPECIAL THERAPEUTIC OR HYGIENIC PURPOSES OR SPECIFIC PARTS OF THE BODY
    • A61H1/00—Apparatus for passive exercising; Vibrating apparatus; Chiropractic devices, e.g. body impacting devices, external devices for briefly extending or aligning unbroken bones
    • A61H1/02—Stretching or bending or torsioning apparatus for exercising
    • A61H1/0237—Stretching or bending or torsioning apparatus for exercising for the lower limbs
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61H—PHYSICAL THERAPY APPARATUS, e.g. DEVICES FOR LOCATING OR STIMULATING REFLEX POINTS IN THE BODY; ARTIFICIAL RESPIRATION; MASSAGE; BATHING DEVICES FOR SPECIAL THERAPEUTIC OR HYGIENIC PURPOSES OR SPECIFIC PARTS OF THE BODY
    • A61H2201/00—Characteristics of apparatus not provided for in the preceding codes
    • A61H2201/12—Driving means
    • A61H2201/1207—Driving means with electric or magnetic drive
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61H—PHYSICAL THERAPY APPARATUS, e.g. DEVICES FOR LOCATING OR STIMULATING REFLEX POINTS IN THE BODY; ARTIFICIAL RESPIRATION; MASSAGE; BATHING DEVICES FOR SPECIAL THERAPEUTIC OR HYGIENIC PURPOSES OR SPECIFIC PARTS OF THE BODY
    • A61H2201/00—Characteristics of apparatus not provided for in the preceding codes
    • A61H2201/16—Physical interface with patient
    • A61H2201/1657—Movement of interface, i.e. force application means
    • A61H2201/1659—Free spatial automatic movement of interface within a working area, e.g. Robot
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61H—PHYSICAL THERAPY APPARATUS, e.g. DEVICES FOR LOCATING OR STIMULATING REFLEX POINTS IN THE BODY; ARTIFICIAL RESPIRATION; MASSAGE; BATHING DEVICES FOR SPECIAL THERAPEUTIC OR HYGIENIC PURPOSES OR SPECIFIC PARTS OF THE BODY
    • A61H2201/00—Characteristics of apparatus not provided for in the preceding codes
    • A61H2201/50—Control means thereof
    • A61H2201/5007—Control means thereof computer controlled
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61H—PHYSICAL THERAPY APPARATUS, e.g. DEVICES FOR LOCATING OR STIMULATING REFLEX POINTS IN THE BODY; ARTIFICIAL RESPIRATION; MASSAGE; BATHING DEVICES FOR SPECIAL THERAPEUTIC OR HYGIENIC PURPOSES OR SPECIFIC PARTS OF THE BODY
    • A61H2205/00—Devices for specific parts of the body
    • A61H2205/06—Arms
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61H—PHYSICAL THERAPY APPARATUS, e.g. DEVICES FOR LOCATING OR STIMULATING REFLEX POINTS IN THE BODY; ARTIFICIAL RESPIRATION; MASSAGE; BATHING DEVICES FOR SPECIAL THERAPEUTIC OR HYGIENIC PURPOSES OR SPECIFIC PARTS OF THE BODY
    • A61H2205/00—Devices for specific parts of the body
    • A61H2205/10—Leg

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Veterinary Medicine (AREA)
  • Public Health (AREA)
  • Epidemiology (AREA)
  • Pain & Pain Management (AREA)
  • Physical Education & Sports Medicine (AREA)
  • Rehabilitation Therapy (AREA)
  • Animal Behavior & Ethology (AREA)
  • Mechanical Engineering (AREA)
  • Robotics (AREA)
  • Manipulator (AREA)

Abstract

The invention provides a rehabilitation robot control method and device and a rehabilitation robot, and relates to the technical field of robot control, wherein the rehabilitation robot control method comprises the following steps: step S102, inputting a training target into the trained neural network model for torque prediction to obtain a torque prediction value of the joint to be trained; step S104, controlling the joint to be trained to move according to a training target based on the moment predicted value; step S106, acquiring and storing actual motion data generated by the joint to be trained, and updating a training data set based on the actual motion data; and S108, updating and training the neural network model based on the updated training data set, and returning the updated and trained neural network model to the S102 as a new trained neural network model. The method and the device can improve the universality of the neural network model and improve the accuracy of the predicted value of the neural network model.

Description

Rehabilitation robot control method and device and rehabilitation robot
Technical Field
The invention relates to the technical field of robot control, in particular to a rehabilitation robot control method and device and a rehabilitation robot.
Background
The existing rehabilitation robot is usually provided with a certain target position when performing rehabilitation training, and assists a part to be trained to reach the set target position, along with the development of deep learning technology, more and more rehabilitation robots use deep learning to solve the training control problem of the rehabilitation robot.
Disclosure of Invention
In view of the above, the present invention provides a rehabilitation robot control method, device and rehabilitation robot, which can improve the universality of a neural network model and improve the accuracy of a predicted value of the neural network model.
In order to achieve the above purpose, the embodiment of the present invention adopts the following technical solutions:
in a first aspect, an embodiment of the present invention provides a rehabilitation robot control method, including: step S102, inputting a training target into the trained neural network model for torque prediction to obtain a torque prediction value of the joint to be trained; step S104, controlling the joint to be trained to move according to the training target based on the moment predicted value; step S106, acquiring and storing actual motion data generated by the joint to be trained, and updating the training data set based on the actual motion data; and S108, updating and training the neural network model based on the updated training data set, and returning the updated and trained neural network model to the S102 as a new trained neural network model.
Further, an embodiment of the present invention provides a first possible implementation manner of the first aspect, where the step of updating the training data set based on the actual motion data includes: judging whether the stored actual motion data reach a preset number or not, and if the actual motion data reach the preset number, determining that the actual motion data meet a preset condition; the actual motion data comprise an actual angle, an actual angular velocity, an actual angular acceleration and an actual moment of the joint to be trained; calculating the sample distance between each sample data in each data group in the training data group and the actual motion data, and replacing the target data group with the maximum sample distance to the actual motion data with the actual motion data to obtain an updated training data group; wherein the number of sample data included in the target data set is the same as the number of actual motion data.
Further, an embodiment of the present invention provides a second possible implementation manner of the first aspect, where the step of performing update training on the neural network model based on the updated training data set includes: randomly extracting a plurality of data sets from the updated training data set as training samples to be input into the neural network model, so that the neural network model carries out moment prediction based on the angle, the angular velocity and the angular acceleration of the joint to be trained in the training samples, and carries out gradient back propagation updating weight based on the error between the predicted moment and the actual moment in the training samples.
Further, an embodiment of the present invention provides a third possible implementation manner of the first aspect, where the neural network model is constructed based on an eulerian lagrange dynamical equation, the neural network model includes a feed-forward network, a first network, a second network, a third network, and a prediction network, an output end of the feed-forward network is connected to input ends of the first network, the second network, and the third network, and output ends of the first network, the second network, and the third network are all connected to the prediction network.
Further, an embodiment of the present invention provides a fourth possible implementation manner of the first aspect, where the training targets include a target angle, a target angular velocity, and a target angular acceleration of the joint to be trained; the step of inputting the training target into the trained neural network model for moment prediction to obtain the moment prediction value of the joint to be trained comprises the following steps: inputting the target angle into the feedforward network, and inputting the target angular velocity and the target angular acceleration into the prediction network, so that the prediction network calculates and outputs the moment prediction value based on the gravity moment vector output by the first network, the diagonal element output by the second network, the lower diagonal element output by the third network, the target angular velocity and the target angular acceleration.
Further, an embodiment of the present invention provides a fifth possible implementation manner of the first aspect, where the calculation formula of the predicted torque value is:
Figure BDA0003362470640000031
wherein, taup(q) is the moment of force predicted value, g (q) is the moment of gravity vector, q is the angle of the joint to be trained,
Figure BDA0003362470640000032
is the angular velocity of the joint to be trained,
Figure BDA0003362470640000033
for the angular acceleration of the joint to be trained, M (q) ═ L (q)TL (q), L (q) is a lower triangular matrix obtained by combining the diagonal elements and the lower diagonal elements,
Figure BDA0003362470640000034
is the moment of the Kelvin term,
Figure BDA0003362470640000035
Figure BDA0003362470640000036
further, an embodiment of the present invention provides a sixth possible implementation manner of the first aspect, where the rehabilitation robot control method further includes: and if the actual motion data does not reach the preset number, returning to repeatedly execute the step S102 to the step S104 until the actual motion data meets the preset condition.
In a second aspect, an embodiment of the present invention further provides a rehabilitation robot control device, including: the using module is used for inputting the training target into the trained neural network model for torque prediction to obtain a torque predicted value of the joint to be trained; the control module is used for controlling the joint to be trained to move according to the training target based on the moment predicted value; the data processing module is used for acquiring and storing actual motion data generated by the joint to be trained and updating the training data set based on the actual motion data; and the training module is used for carrying out updating training on the neural network model based on the updated training data set and transmitting the updated and trained neural network model to the using module as a new trained neural network model.
In a third aspect, an embodiment of the present invention provides a rehabilitation robot, including: a processor and a storage device; the storage means has stored thereon a computer program which, when executed by the processor, performs the method of any of the first aspects.
In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, performs the steps of the method according to any one of the above first aspects.
The embodiment of the invention provides a rehabilitation robot control method, a rehabilitation robot control device and a rehabilitation robot, wherein the rehabilitation robot control method comprises the following steps: step S102, inputting a training target into the trained neural network model for torque prediction to obtain a torque prediction value of the joint to be trained; step S104, controlling the joint to be trained to move according to a training target based on the moment predicted value; step S106, acquiring and storing actual motion data generated by the joint to be trained, and updating a training data set based on the actual motion data; and S108, updating and training the neural network model based on the updated training data set, and returning the updated and trained neural network model to the S102 as a new trained neural network model. In the rehabilitation robot control method provided by this embodiment, in the process of training and controlling the joint to be trained based on the neural network model, the training data set is updated based on the actual motion data of the joint to be trained, and the neural network model is trained on line based on the updated training data set, so that the universality of the neural network model is improved, and the accuracy of the predicted value of the neural network model is improved.
Additional features and advantages of embodiments of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of embodiments of the invention as set forth above.
In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
Fig. 1 is a flowchart illustrating a control method for a rehabilitation robot according to an embodiment of the present invention;
FIG. 2 is a schematic diagram illustrating a training data set update provided by an embodiment of the present invention;
FIG. 3 is a flow chart illustrating the use of a neural network model provided by an embodiment of the present invention;
FIG. 4 is a schematic diagram illustrating a neural network model moment prediction provided by an embodiment of the present invention;
FIG. 5 is a flow chart illustrating a neural network model training process provided by an embodiment of the present invention;
fig. 6 shows a flowchart of a control of a rehabilitation robot according to an embodiment of the present invention;
fig. 7 shows a schematic structural diagram of a rehabilitation robot control device according to an embodiment of the invention.
Detailed Description
To make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some, not all, embodiments of the present invention.
There are two well-known problems in robot dynamics control, the positive and negative. The positive problem is that the driving force of a motor at each joint of the robot is known, and the final arriving position, momentum and acceleration of the mechanical arm are predicted; the inverse problem is how to distribute the driving force of the motors at each joint of the robot in order to move the mechanical arm to a specified position and have a specified momentum or acceleration. At present, a common treatment scheme of the upper limb rehabilitation robot is to set a certain target position and require the limbs of a patient to reach the target position with the assistance of the robot, namely the inverse problem of the robot. With the development of deep learning technology, more and more recovery robots choose to use deep learning (neural network) to solve the control problem, and at present, a common method is to use historical data to train a neural network model offline first, and then use the pre-trained fixed neural network to perform training control, and the control method has the following problems:
(1) the neural network model learned offline cannot completely characterize the characteristics of the robot system, resulting in insufficient universality. The offline data set is limited, the robot can continuously generate new data during actual work, and the model trained offline can not predict the new data; the off-line model assumes that the data distribution is fixed or stable, the training samples are independently and uniformly distributed, but the robot may cause the data distribution to change continuously due to the change of the working conditions during the actual work, and the distribution of the training data is not stable in correlation.
(2) Due to the randomness of the neural network, results that do not satisfy the physical laws (momentum conservation, energy conservation, etc.) are easily obtained using conventional neural network models. Even if the training data meet all physical laws, the trained neural network model still can make non-physical prediction, and the control accuracy of the rehabilitation robot is reduced.
In order to solve the above problems, embodiments of the present invention provide a rehabilitation robot control method, apparatus, and rehabilitation robot, which can be applied to improve the universality of a neural network model and improve the accuracy of a predicted value of the neural network model. The following describes embodiments of the present invention in detail.
The present embodiment provides a rehabilitation robot control method, which may be applied to a controller of a rehabilitation robot, and referring to a flowchart of the rehabilitation robot control method shown in fig. 1, the method mainly includes the following steps S102 to S106:
and S102, inputting the training target into the trained neural network model for moment prediction to obtain a moment prediction value of the joint to be trained.
The training target comprises a target angle, a target angular velocity and a target angular acceleration of each joint to be trained. The trained neural network model is stored in the using module, and comprises a neural network model obtained by training based on a historical training data set for the first time and a neural network model obtained by updating and training.
The rehabilitation robot may be an upper limb rehabilitation robot, and the training target may be input by the user, or may be calculated by inverse kinematics based on a target position of the tip of the upper limb input by the user and the current position. When receiving the training target, the target values (recorded as target angle, target angular velocity and target angular acceleration) of the angle, angular velocity and angular acceleration vector of n joints to be trained of the rehabilitation robot are recorded
Figure BDA0003362470640000071
Inputting the target to the trained neural network model, and outputting predicted values of moments tau of n joints to be trained by the neural network modelp. Due to the fact that
Figure BDA0003362470640000072
And
Figure BDA0003362470640000073
can be represented by qdIs derived, so that, in practical applications, it may only be necessary to give the target angle q of the joint to be trainedd。
And step S104, controlling the joint to be trained to move according to the training target based on the moment predicted value.
After the neural network model obtains a moment predicted value based on the prediction of a training target, a moment instruction is sent to each joint to be trained based on the moment predicted value, the moment is applied to the joint to be trained, so that the joint to be trained carries out rehabilitation training according to the training target, and actual motion data are generated
Figure BDA0003362470640000074
And step S106, acquiring and storing actual motion data generated by the joint to be trained, and updating the training data set based on the actual motion data.
And storing the actual motion data generated by the joint to be trained each time, and updating the training data set based on the generated new actual motion data. In one possible embodiment, the training data set is updated in real time based on the actual movement data generated by the joint to be trained in order to retrain the neural network model, i.e. one for each generation
Figure BDA0003362470640000075
It is substituted for the historical motion data in the training data set that was generated at the earliest time or the farthest distance from its sample.
In another possible embodiment, it may be determined whether the stored actual exercise data satisfies a preset condition, and when the stored actual exercise data satisfies the preset condition, the training data set may be updated based on the stored actual exercise data. The preset condition may be that the actual motion data meets a certain data volume requirement or belongs to data generated within a preset time length.
In practical application, the training data set can be stored in a data processing moduleFor example, the training data set includes P sets of data generated prior to the joint to be trained (A for each set, each data set being
Figure BDA0003362470640000076
) And historical motion data, wherein when the actual motion data generated by the joint to be trained is full of one group A, the data processing module replaces any one of P groups of historical motion data of the training data group with the actual motion data.
And S108, updating and training the neural network model based on the updated training data set, and returning the updated and trained neural network model to the S102 as a new trained neural network model.
The data processing module sends the updated training data set to the training module to retrain the neural network model, the training module sends the retrained neural network model to the using module (i.e. the retrained neural network model is sent to the using module to be used as the new trained neural network model in the step S102), when the rehabilitation robot receives a new training task, the step S102 is executed in a returning mode, and torque prediction is carried out based on the retrained neural network model until training is finished.
In practical applications, the moment prediction based on the neural network model in step S102 and the retraining of the neural network model in step S108 may be performed simultaneously and in parallel based on different processors, so as to implement online updating of the neural network model.
In the rehabilitation robot control method provided by this embodiment, in the process of training and controlling the joint to be trained based on the neural network model, the training data set is updated based on the actual motion data of the joint to be trained, and the neural network model is trained on line based on the updated training data set, so that the universality of the neural network model is improved, and the accuracy of the predicted value of the neural network model is improved.
In a possible implementation manner, the embodiment provides an implementation manner that determines whether the stored actual exercise data meets the preset condition, and if so, updates the training data set based on the actual exercise data, which may be specifically executed with reference to the following steps (1) to (2):
step (1): and judging whether the stored actual motion data reach the preset number or not, and if the actual motion data reach the preset number, determining that the actual motion data meet the preset condition.
The actual motion data comprises an actual angle, an actual angular velocity, an actual angular acceleration and an actual moment of the joint to be trained. And when the number of the stored actual movement data is the same as the number of data in each data group in the training data group (the number of the stored actual movement data reaches A), determining that the actual movement data meets the preset condition.
Step (2): and calculating the sample distance between each sample data in each data group in the training data group and the actual motion data, and replacing the target data group with the maximum sample distance from the actual motion data with the actual motion data to obtain an updated training data group.
The target data set includes the same number of sample data as the number of actual motion data. The data processing module is stored with a training data group, the training data group comprises P groups of historical motion data, and meanwhile, actual motion data generated by actual motion of the rehabilitation robot is read and stored in real time
Figure BDA0003362470640000091
When the actual motion data reaches a number (a) ((
Figure BDA0003362470640000092
As one data), one of the training data sets having the largest sample distance from the actual motion data is replaced with the newly generated actual motion data.
The sample distance can be expressed using a norm, and the algorithm pseudo-code is as follows:
max＝0；
index＝0；
for i＝1:P
Figure BDA0003362470640000093
ifd(i)>max
max＝d(i)；
index＝i；
end
end
return index
each sample data in the training data group is a pair of n-dimensional vectors measured at the same sampling time
Figure BDA0003362470640000094
q is typically measured by the robot joint encoders,
Figure BDA0003362470640000095
and the t is obtained by q differentiation, and is obtained by measuring each joint torque sensor or converting each joint current value. Alpha is alpha1,α2,α3，α4The angle, the angular velocity, the angular acceleration and the weight of the moment of the joint to be trained, respectively. The data processing module stores the training data set in a linked list mode, and when data replacement occurs, the replacement can be quickly completed only by modifying the linked list pointer.
In a feasible implementation manner, if the actual motion data does not meet the preset condition, the step S102 to the step S104 are repeatedly executed to enable the trained neural network model to continue to perform moment prediction, and the training joints are controlled to move according to the training target to generate new actual motion data until the actual motion data meets the preset condition, that is, the actual motion data reaches the preset number, and the training data set is updated to retrain the neural network model.
Referring to the schematic diagram of updating the training data set shown in fig. 2, a sample K in fig. 2 is a data set to be replaced in the training data set, when replacing, only links between the sample K and a front node K-1 and a rear node K +1 need to be disconnected, then the front node K-1 points to a new sample (i.e., actual motion data), and the new sample points to the K +1 sample to complete the replacement, so as to obtain the updated training data set. And sending the updated training data set to a training module, and retraining the training module once every time the replacement is completed, so that the training frequency of the neural network model can be adjusted by setting the number A of samples in each data in the training data set.
In a feasible implementation manner, the neural network model is constructed based on an eulerian lagrange kinetic equation, the neural network model comprises a feed-forward network, a first network, a second network, a third network and a prediction network, an output end of the feed-forward network is respectively connected with input ends of the first network, the second network and the third network, and output ends of the first network, the second network and the third network are respectively connected with the prediction network.
Referring to the neural network model usage flow chart shown in fig. 3, the above-mentioned feedforward network 30 includes an input layer (dimension n × m), a LeakyReLu layer 1, a fully-connected layer 1 (dimension m × m), a LeakyReLu layer 2, a fully-connected layer 2 (dimension m × m), and a LeakyReLu layer 3; the first network 31 (may also be referred to as a g network) includes a fully connected layer 3 (dimension m × n), the second network 32 (may also be referred to as an Ld network) includes a fully connected layer 4 (dimension m × n) and a SoftPlus layer, and the third network 33 (may also be referred to as a Lo network) includes a fully connected layer 5 (dimension m × n)
Figure BDA0003362470640000111
) (ii) a The prediction network comprises a moment prediction model constructed on the basis of an Eulerian Lagrange kinetic equation.
As shown in fig. 3, the angles (dimension n × 1) of n joints to be trained of the rehabilitation robot are input into the feedforward network 30, sequentially pass through the input layer (dimension n × m), the leak ReLu layer 1, the fully connected layer 1 (dimension m × m), the leak ReLu layer 2, the fully connected layer 2 (dimension m × m), and the leak ReLu layer 3, and output m × 1-dimensional intermediate vectors respectively enter the first network, the second network, and the third network (i.e., the g network, the Ld network, and the Lo network). The first network comprises only the fully-connected layer 3 (dimension m × n), outputting the gravitational moment terms g (q); the second network comprises a fully connected layer 4 (dimension m × n) and a SoftPlus layer, outputting a diagonal element Ld(q) selecting SoftPlus as the guaranteed output L of the activation functiondEach term of (q) is greater than zero; the third network contains only fully connected layer 5 (dimension)
Figure BDA0003362470640000112
) Output lower diagonal element Lo(q)。
In a feasible implementation manner, this embodiment provides the specific implementation manner that the training target is input into the trained neural network model for moment prediction to obtain the predicted moment value of the joint to be trained: inputting a target angle of a training target into a feedforward network, and inputting a target angular velocity and a target angular acceleration into a prediction network, so that the prediction network calculates and outputs a moment prediction value based on a gravity moment vector output by a first network, a diagonal element output by a second network, a lower diagonal element output by a third network, the target angular velocity and the target angular acceleration.
When the moment prediction is performed on the training target based on the trained neural network model, referring to the moment prediction schematic diagram of the neural network model shown in fig. 4, the target angle q of the joint to be trained in the training target is determineddInputting into a feedforward network, making the first network output a gravity moment vector g (q)d) Into the prediction network, the second network outputs a diagonal element Ld(q) into a prediction network, wherein the diagonal elements Ld(q) n; diagonal element L at third network outputo(q) lower diagonal element L into the prediction networko(q) is as follows
Figure BDA0003362470640000113
And (4) respectively. Predicting the network will diagonal element Ld(q) lower diagonal element Lo(q) combining to obtain a lower triangular matrix L (q) (L (q)) epsilon Rn×nIs a lower triangular matrix), the inertia matrix m (q) is a positive definite symmetric matrix, so the cholesky decomposition can be used to construct the inertia matrix m (q) ═ l (q)TAnd L (q), the construction mode can ensure that the obtained inertia matrix meets the positive definite symmetry characteristic (namely, the mass can only be larger than zero but not smaller than or equal to zero), and the current traditional neural network structure cannot ensure the characteristic.
Mixing M (q) with the angular acceleration of the joint to be trained
Figure BDA0003362470640000121
Multiplying to obtain moment of inertia term
Figure BDA0003362470640000122
Computing the moment of the Kelvin term and calculating the moment of the inertia term
Figure BDA0003362470640000123
Coco type term moment
Figure BDA0003362470640000124
And the three items of gravity moment g (q) are added to obtain a predicted value tau of the joint output moment vectorp。
The calculation formula of the moment predicted value is as follows:
Figure BDA0003362470640000125
wherein, taupIs a moment predicted value, g (q) is a gravity moment vector, q is an angle of a joint to be trained,
Figure BDA0003362470640000126
in order to determine the angular velocity of the joint to be trained,
Figure BDA0003362470640000127
for the angular acceleration of the joint to be trained, M (q) ═ L (q)TL (q), L (q) is a lower triangular matrix obtained by combining the diagonal elements and the lower diagonal elements,
Figure BDA0003362470640000128
is the moment of the Kelvin term,
Figure BDA0003362470640000129
*Trepresenting a transpose of a matrix or vector.
Target angle q of the training targetdTarget angular velocity
Figure BDA00033624706400001210
And target angular acceleration
Figure BDA00033624706400001211
Correspondingly inputting the calculation formula of the moment predicted value to obtain the moment predicted value
Figure BDA00033624706400001212
Figure BDA00033624706400001213
The above target angle qdTarget angular velocity
Figure BDA00033624706400001214
And target angular acceleration
Figure BDA00033624706400001215
Are all vectors.
In a possible implementation, the present embodiment provides a specific implementation of the update training of the neural network model based on the updated training data set: and randomly extracting a plurality of data sets from the updated training data set as training samples to be input into the neural network model, so that the neural network model carries out moment prediction based on the angle, the angular velocity and the angular acceleration of the joint to be trained in the training samples, and carries out gradient back propagation updating on the weight based on the error between the moment obtained by prediction and the actual moment in the training samples.
The training module randomly samples Q data from the P × a data of the updated training data set as training samples, inputs the training samples into the neural network model, and retrains the neural network model.
Referring to the neural network model training flowchart shown in FIG. 5, the training samples are
Figure BDA0003362470640000131
Inputting the data into a neural network model, wherein,
Figure BDA0003362470640000132
respectively representing the angle, the angular velocity and the angular acceleration vector of n joints to be trained of the rehabilitation robot, and tau belongs to RnRepresenting n joints to be trained
Figure BDA0003362470640000133
As shown in fig. 5, the actual value of the historical output torque vector, i.e., the angle q of the joint to be trained, is input into a feedforward neural network, and the gravity torque g (q) and the diagonal element L are respectively output through a first network, a second network and a third networkd(q) and lower diagonal element Lo(q) reacting Ld(q) and Lo(q) combining to obtain a lower triangular matrix L (q), and then M (q) ═ L (q)TL (q) calculating M (q), and calculating M (q) and the angular acceleration of the joint
Figure BDA0003362470640000134
Multiplying to obtain moment of inertia term
Figure BDA0003362470640000135
Computing the moment of the Kelvin term and calculating the moment of the inertia term
Figure BDA0003362470640000136
Coco type term moment
Figure BDA0003362470640000137
And adding three gravity moment terms g (q) to obtain a moment predicted value tau of the joint output moment vectorp，
Figure BDA0003362470640000138
Figure BDA0003362470640000139
Finally, predicting the value tau through the momentpMean square error with actual value tau of historical output torque vector of joint to be trained in training sample
Figure BDA00033624706400001310
Gradient backpropagation is performed to update the weights of the neural network model.
According to the control method of the rehabilitation robot, the actual motion data generated by the joint to be trained is obtained and processed, so that the training and the use of the neural network are executed in parallel, the control speed of the rehabilitation robot is increased, the online updating of the neural network model is realized, the universality of the neural network model is improved, the structure of the neural network is constructed based on the Euler Lagrange kinetic equation, the prediction result is ensured to accord with the physical law, and the control accuracy and the reliability of the rehabilitation robot are improved.
On the basis of the foregoing embodiment, the present embodiment provides an example of performing training control on a rehabilitation robot (such as the upper limb rehabilitation exoskeleton robot Wisebot X5) by applying the foregoing rehabilitation robot control method, which can be specifically executed with reference to the following steps 1 to 6:
step 1, loading P groups (A in each group) of historical data
Figure BDA0003362470640000141
Loading the neural network model obtained by pre-training on a data processing module, and giving a rehabilitation training target
Figure BDA0003362470640000142
Referring to the control flow chart of the rehabilitation robot shown in fig. 6, the neural network model and the rehabilitation training target obtained by pre-training are combined
Figure BDA0003362470640000143
And loading the data into the use model so that the use model can carry out moment prediction based on the neural network model. The training module is a group of graphic processors (GPU A), the using module is another group of graphic processors (GPU B), the GPU A is used for training the neural network, the GPU B is used for using the neural network, and the GPU A and the GPU B can be executed in parallel to realize online updating of the neural network.
Step 2, reading the rehabilitation training target by using the module
Figure BDA0003362470640000144
Inputting the moment predicted value tau of each joint to be trained in the neural network modelPAnd sent to each joint to be trained.
Step 3, each joint to be trained receives a torque command tauPApplying torque to the joint to be trained to enable the joint to be trained to perform rehabilitation training movement according to the target, and generating actual movement data
Figure BDA0003362470640000145
Step 4, actual motion data sample
Figure BDA0003362470640000146
And when the number of the groups is A, the data processing module replaces one group in P groups of historical motion data in the training data group, and sends new P groups of historical operation data to the training module.
The data processing module processes a continuously input data stream (actual motion data generated by a joint to be trained) based on empirical replay, wherein the empirical replay refers to the mixed use of new data and old data when training a neural network model, a part of new data representing the latest state of the robot is used for model training, and a part of representative old data is reserved and used for model training. The training data set is divided by taking a set as a unit, the data processing module stores P sets of data, each set of data comprises A (one historical motion data comprises the angle, the angular velocity, the angular acceleration and the joint moment of a joint to be trained), and the training data set is stored in a linked list mode.
And 5, the training module randomly samples Q from the new P × A historical motion data to be used as training samples to retrain the neural network model, and sends the new neural network model to the use module for use after the updating and training are completed.
And during training, the neural network model randomly samples Q data from P x A data stored in the data processing module to serve as training samples, and divides the training samples into a training set, a verification set and a test set. Neural network model input rehabilitation robot n angles and angles of joints to be trainedActual values of velocity, angular acceleration and moment vectors
Figure BDA0003362470640000151
Outputting predicted values tau of moment vectors of n joints to be trained of rehabilitation robotpAnd by predicting τpThe mean square error with the actual value τ is propagated back gradiently to update the weights of the neural network.
Step 6, updating the neural network model by using the module and reading the next rehabilitation training target
Figure BDA0003362470640000152
And (5) circularly executing the steps 2 to 6 until the rehabilitation training is finished.
Corresponding to the rehabilitation robot control method provided in the above embodiment, an embodiment of the present invention provides a rehabilitation robot control device, which is applied to a controller of a rehabilitation robot, and referring to a schematic structural diagram of a rehabilitation robot control device shown in fig. 7, the device includes the following modules:
the using module 71 is used for inputting a training target into the trained neural network model for torque prediction to obtain a torque predicted value of the joint to be trained;
the control module 72 is used for controlling the joint to be trained to move according to a training target based on the moment predicted value;
the data processing module 73 is used for acquiring and storing actual motion data generated by the joint to be trained and updating the training data set based on the actual motion data;
and the training module 74 is configured to perform update training on the neural network model based on the updated training data set, and transmit the updated and trained neural network model to the use module as a new trained neural network model.
According to the rehabilitation robot control device provided by the embodiment, the training data set is updated based on the actual motion data of the joint to be trained in the process of training and controlling the joint to be trained based on the neural network model, and the neural network model is trained on line based on the updated training data set, so that the universality of the neural network model is improved, and the accuracy of the predicted value of the neural network model is improved.
In an embodiment, the data processing module 73 is further configured to determine whether the stored actual motion data reaches a preset number, and if the stored actual motion data reaches the preset number, determine that the actual motion data meets a preset condition; the actual motion data comprises an actual angle, an actual angular velocity, an actual angular acceleration and an actual moment of the joint to be trained; calculating the sample distance between each sample data in each data group in the training data group and the actual motion data, and replacing the target data group with the maximum sample distance to the actual motion data with the actual motion data to obtain an updated training data group; the number of sample data included in the target data group is the same as the number of actual motion data.
In an embodiment, the training module 74 is further configured to randomly extract a plurality of data sets from the updated training data set as training samples to be input into the neural network model, so that the neural network model performs torque prediction based on the angle, the angular velocity, and the angular acceleration of the joint to be trained in the training samples, and performs gradient back propagation to update the weight based on an error between the predicted torque and the actual torque in the training samples.
In an embodiment, the neural network model is constructed based on an eulerian lagrange kinetic equation, the neural network model includes a feed-forward network, a first network, a second network, a third network and a prediction network, an output end of the feed-forward network is connected with input ends of the first network, the second network and the third network respectively, and output ends of the first network, the second network and the third network are connected with the prediction network.
In one embodiment, the training target includes a target angle, a target angular velocity and a target angular acceleration of the joint to be trained; the using module 71 is further configured to input the target angle into the feedforward network, and input the target angular velocity and the target angular acceleration into the prediction network, so that the prediction network calculates and outputs a predicted moment value based on the gravity moment vector output by the first network, the diagonal element output by the second network, the lower diagonal element output by the third network, the target angular velocity, and the target angular acceleration.
In one embodiment, the calculation formula of the predicted torque value is:
Figure BDA0003362470640000161
wherein, taupIs a moment predicted value, g (q) is a gravity moment vector, q is an angle of a joint to be trained,
Figure BDA0003362470640000162
in order to determine the angular velocity of the joint to be trained,
Figure BDA0003362470640000163
for the angular acceleration of the joint to be trained, M (q) ═ L (q)TL (q), L (q) is a lower triangular matrix obtained by combining diagonal elements and lower diagonal elements,
Figure BDA0003362470640000171
is the moment of the Kelvin term,
Figure BDA0003362470640000172
in an embodiment, the data processing module 73 is further configured to return to the using module and the data processing module until the actual exercise data does not reach the preset number, until the actual exercise data meets the preset condition.
According to the control device of the rehabilitation robot, the actual motion data generated by the joint to be trained is obtained and processed, so that the training and the use of the neural network are executed in parallel, the control speed of the rehabilitation robot is increased, the online updating of the neural network model is realized, the universality of the neural network model is improved, the structure of the neural network is constructed based on the Euler Lagrange kinetic equation, the prediction result is ensured to accord with the physical law, and the control accuracy and the reliability of the rehabilitation robot are improved.
The device provided by the embodiment has the same implementation principle and technical effect as the foregoing embodiment, and for the sake of brief description, reference may be made to the corresponding contents in the foregoing method embodiment for the portion of the embodiment of the device that is not mentioned.
Corresponding to the method and the device provided by the foregoing embodiments, an embodiment of the present invention further provides a rehabilitation robot, including: an upper extremity exoskeleton robot and a controller, the controller comprising a processor and a memory device; the storage device stores thereon a computer program which, when executed by the processor, executes the rehabilitation robot control method provided by the above-described embodiment.
Embodiments of the present invention provide a computer-readable medium, wherein the computer-readable medium stores computer-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the method of the above-mentioned embodiments.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working process of the system described above may refer to the corresponding process in the foregoing embodiments, and is not described herein again.
The rehabilitation robot control method, the rehabilitation robot control device and the computer program product of the rehabilitation robot provided by the embodiment of the invention comprise a computer readable storage medium storing program codes, instructions included in the program codes can be used for executing the method in the previous method embodiment, and specific implementation can refer to the method embodiment and is not described herein again.
In addition, in the description of the embodiments of the present invention, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, e.g., as meaning either a fixed connection, a removable connection, or an integral connection; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of description and simplicity of description, but do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
Finally, it should be noted that: the above-mentioned embodiments are only specific embodiments of the present invention, which are used for illustrating the technical solutions of the present invention and not for limiting the same, and the protection scope of the present invention is not limited thereto, although the present invention is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit and scope of the embodiments of the present invention, and they should be construed as being included therein. Therefore, the protection scope of the present invention shall be subject to the protection scope of the appended claims.

Claims (10)

1. A rehabilitation robot control method is characterized by comprising the following steps:
step S102, inputting a training target into the trained neural network model for torque prediction to obtain a torque prediction value of the joint to be trained;
step S104, controlling the joint to be trained to move according to the training target based on the moment predicted value;
step S106, acquiring and storing actual motion data generated by the joint to be trained, and updating the training data set based on the actual motion data;
and S108, updating and training the neural network model based on the updated training data set, and returning the updated and trained neural network model to the S102 as a new trained neural network model.
2. The method of claim 1, wherein the step of updating the training data set based on the actual motion data comprises:
judging whether the stored actual motion data reach a preset number or not, and if the actual motion data reach the preset number, determining that the actual motion data meet a preset condition; the actual motion data comprise an actual angle, an actual angular velocity, an actual angular acceleration and an actual moment of the joint to be trained;
calculating the sample distance between each sample data in each data group in the training data group and the actual motion data, and replacing the target data group with the maximum sample distance to the actual motion data with the actual motion data to obtain an updated training data group; wherein the number of sample data included in the target data set is the same as the number of actual motion data.
3. The method of claim 1, wherein the step of training the neural network model based on the updated training dataset comprises:
randomly extracting a plurality of data sets from the updated training data set as training samples to be input into the neural network model, so that the neural network model carries out moment prediction based on the angle, the angular velocity and the angular acceleration of the joint to be trained in the training samples, and carries out gradient back propagation updating weight based on the error between the predicted moment and the actual moment in the training samples.
4. The method according to claim 1, wherein the neural network model is constructed based on eulerian Lagrange dynamics equations, the neural network model comprises a feed forward network, a first network, a second network, a third network and a prediction network, an output end of the feed forward network is connected with an input end of the first network, an input end of the second network and an input end of the third network, and output ends of the first network, the second network and the third network are connected with the prediction network.
5. The method of claim 4, wherein the training goals comprise a target angle, a target angular velocity, and a target angular acceleration of the joint to be trained;
the step of inputting the training target into the trained neural network model for moment prediction to obtain the moment prediction value of the joint to be trained comprises the following steps:
inputting the target angle into the feedforward network, and inputting the target angular velocity and the target angular acceleration into the prediction network, so that the prediction network calculates and outputs the moment prediction value based on the gravity moment vector output by the first network, the diagonal element output by the second network, the lower diagonal element output by the third network, the target angular velocity and the target angular acceleration.
6. The method of claim 5, wherein the moment prediction value is calculated by:
Figure FDA0003362470630000021
wherein, taup(q) is the moment of force predicted value, g (q) is the moment of gravity vector, q is the angle of the joint to be trained,
Figure FDA0003362470630000022
is the angular velocity of the joint to be trained,
Figure FDA0003362470630000023
for the angular acceleration of the joint to be trained, M (q) ═ L (q)TL (q), L (q) is a lower triangular matrix obtained by combining the diagonal elements and the lower diagonal elements,
Figure FDA0003362470630000031
is the moment of the Kelvin term,
Figure FDA0003362470630000032
Figure FDA0003362470630000033
7. the method of claim 2, further comprising:
and if the actual motion data does not reach the preset number, returning to repeatedly execute the step S102 to the step S104 until the actual motion data meets the preset condition.
8. A rehabilitation robot control device characterized by comprising:
the using module is used for inputting the training target into the trained neural network model for torque prediction to obtain a torque predicted value of the joint to be trained;
the control module is used for controlling the joint to be trained to move according to the training target based on the moment predicted value;
the data processing module is used for acquiring and storing actual motion data generated by the joint to be trained and updating the training data set based on the actual motion data;
and the training module is used for carrying out updating training on the neural network model based on the updated training data set and transmitting the updated and trained neural network model to the using module as a new trained neural network model.
9. A rehabilitation robot, comprising: a processor and a storage device;
the storage device has stored thereon a computer program which, when executed by the processor, performs the method of any one of claims 1 to 7.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method according to any one of the claims 1 to 7.
CN202111371350.5A 2021-11-18 2021-11-18 Rehabilitation robot control method and device and rehabilitation robot Pending CN114028164A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111371350.5A CN114028164A (en) 2021-11-18 2021-11-18 Rehabilitation robot control method and device and rehabilitation robot

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111371350.5A CN114028164A (en) 2021-11-18 2021-11-18 Rehabilitation robot control method and device and rehabilitation robot

Publications (1)

Publication Number Publication Date
CN114028164A true CN114028164A (en) 2022-02-11

Family

ID=80138158

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111371350.5A Pending CN114028164A (en) 2021-11-18 2021-11-18 Rehabilitation robot control method and device and rehabilitation robot

Country Status (1)

Country Link
CN (1) CN114028164A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023245937A1 (en) * 2022-06-21 2023-12-28 郑州安杰莱智能科技有限公司 Conversion method and apparatus for current moment, and rehabilitation robot

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109732605A (en) * 2019-01-21 2019-05-10 厦门大学 A kind of compensation method and system of joint of robot moment of friction
CN111888194A (en) * 2020-08-06 2020-11-06 广州一康医疗设备实业有限公司 Upper limb rehabilitation training method, system, device and storage medium
CN112247992A (en) * 2020-11-02 2021-01-22 中国科学院深圳先进技术研究院 Robot feedforward torque compensation method
CN112540620A (en) * 2020-12-03 2021-03-23 西湖大学 Reinforced learning method and device for foot type robot and electronic equipment

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109732605A (en) * 2019-01-21 2019-05-10 厦门大学 A kind of compensation method and system of joint of robot moment of friction
CN111888194A (en) * 2020-08-06 2020-11-06 广州一康医疗设备实业有限公司 Upper limb rehabilitation training method, system, device and storage medium
CN112247992A (en) * 2020-11-02 2021-01-22 中国科学院深圳先进技术研究院 Robot feedforward torque compensation method
CN112540620A (en) * 2020-12-03 2021-03-23 西湖大学 Reinforced learning method and device for foot type robot and electronic equipment

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023245937A1 (en) * 2022-06-21 2023-12-28 郑州安杰莱智能科技有限公司 Conversion method and apparatus for current moment, and rehabilitation robot

Similar Documents

Publication Publication Date Title
US20230330848A1 (en) Reinforcement and imitation learning for a task
CN108621159B (en) Robot dynamics modeling method based on deep learning
CN112904728B (en) Mechanical arm sliding mode control track tracking method based on improved approach law
Qazani et al. Optimising control and prediction horizons of a model predictive control-based motion cueing algorithm using butterfly optimization algorithm
Hafez et al. Deep intrinsically motivated continuous actor-critic for efficient robotic visuomotor skill learning
CN112571420B (en) Dual-function model prediction control method under unknown parameters
CN116460860B (en) Model-based robot offline reinforcement learning control method
Perrusquía et al. Nonlinear control using human behavior learning
CN114028164A (en) Rehabilitation robot control method and device and rehabilitation robot
Jiang et al. Generative adversarial interactive imitation learning for path following of autonomous underwater vehicle
US20220058472A1 (en) Implementing a multi-layer neural network using a single physical layer of analog neurons
CN116638507A (en) Teleoperation control method combining self-adaptive impedance control and predictive control
CN113650014B (en) Redundant mechanical arm tracking control method based on echo state network
CN108927806A (en) A kind of industrial robot learning method applied to high-volume repeatability processing
CN112292239B (en) Method and device for computer-aided determination of control parameters for a suitable operating technical system
CN114118371A (en) Intelligent agent deep reinforcement learning method and computer readable medium
Bedolla-Martinez et al. Learning human inverse kinematics solutions for redundant robotic upper-limb rehabilitation
CN116852338A (en) Method and device for predicting upper limb movement pose of hanging rod exoskeleton
Hassanein et al. Auto-generating fuzzy system modelling of physical systems
Zuo et al. Flight attitude simulator control system design based on model-free reinforcement learning method
Mu et al. A Data-Based Feedback Relearning Algorithm for Uncertain Nonlinear Systems
CN115319741B (en) Robot control model training method and robot control method
Tao et al. An Efficient Learning Control Framework With Sim-to-Real for String-Type Artificial Muscle-Driven Robotic Systems
Kumarawadu et al. Neural network‐based optimal adaptive tracking using genetic algorithms
Guo et al. Application of particle swarm optimisation algorithm in manipulator compliance control

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