KR20220039205A - System and method for controlling multi-degree-of-freedom robot - Google Patents

Abstract

The present invention relates to a system and a method for controlling a multi-degree-of-freedom robot, which can calculate a joint angle for numerical analysis for inverse kinematic analysis by using a deep learning algorithm, and move a tool or an end of a multi-degree-of-freedom robot to a target position by using the joint angle for numerical analysis. Also, the present invention (1) generates learning data including a plurality of joint angles for learning, reaching positions for learning which are positions reached by a tool or an end of a multi-degree-of-freedom robot in accordance with the joint angles for learning, (2) uses the learning data to learn joint angles for numerical analysis corresponding to the position of the tool or the end of the multi-degree-of-freedom robot, (3) calculates at least one joint angle for numerical analysis corresponding to a target position to which the tool or the end of the multi-degree-of-freedom robot is to be moved when the target position is inputted, and (4) calculates a target joint angle at which the tool or the end of the multi-degree-of-freedom robot can reach the target position from at least one joint angle for numerical analysis through inverse kinematic analysis to reduce the burden of a developer who performs a kinematic-based calculation process and resolve limitations on the number, form, and degree of freedom of joints provided on the multi-degree-of-freedom robot.

Description

Multi-degree-of-freedom robot control method and system

The present invention calculates a joint angle for numerical analysis for inverse kinematics analysis using a deep learning algorithm, and a target joint angle that can move the distal end or tool of a multi-DOF robot to a target position using the joint angle for numerical analysis It relates to a method and system for controlling a multi-degree-of-freedom robot for calculating .

In general, a multiple degree of freedom robot having a plurality of joints must rotate by a predetermined angle by driving a motor or actuator provided in each joint in order to move a distal end or a tool to a target position.

Therefore, when information about the target position is given, the controller or control system provided in the multi-degree-of-freedom robot calculates the joint angles of the joints in real time from information about the coordinate values and directions of the target position, or is calculated in advance by the developer. The motor or actuator is controlled by extracting the stored joint angle.

On the other hand, in order to calculate the joint angle from the coordinate values and direction information of the target position in the controller or control system provided in the multi-DOF robot, an inverse kinematics-based calculation process must be performed.

This inverse kinematics-based calculation process is performed by performing inverse kinematics analysis while updating arbitrary values corresponding to joint angles, and this process is repeatedly performed until a joint angle corresponding to the target position is calculated.

That is, in the prior art, a large amount of calculation time and amount of calculation are required to calculate the joint angle corresponding to the target position in the controller or control system provided in the multi-DOF robot, and the arbitrary value substituted for the first time for inverse kinematics analysis is If there is a significant difference from the joint angle corresponding to the target position, there is a problem in that the analysis of the joint angle corresponding to the target position is impossible or the calculation time is further delayed.

In addition, even when the developer wants to calculate and store the joint angle corresponding to the target position in advance, the regular kinematics-based calculation process for calculating the target position from the joint angle of the multi-DOF robot is DH parameters (Denavit-Hartenberg parameters, DH parameters) can be easily calculated in a relatively standardized way, but since there is no standardized method for the inverse kinematics-based calculation process that calculates the joint angle of the multi-DOF robot from the target position, the developer has to work Since it is necessary to calculate the joint angle corresponding to the target position at once, there is a problem in that the burden on the developer is increased and the accuracy of the joint angle is lowered.

In addition, in order to facilitate the inverse kinematics-based calculation process, developers of multi-DOF robots frequently limit the number, shape, and degree of freedom of joints provided in the multi-DOF robot.

The present invention has been devised in view of the above problems, and an object of the present invention is to provide a method and system for controlling a multi-DOF robot capable of controlling the multi-DOF robot by calculating the joint angle of the multi-DOF robot from a target position. there is.

Another object of the present invention is to provide a method and system for controlling a multi-DOF robot capable of calculating joint angles without limitation due to the joint configuration of the multi-DOF robot.

Objects of the present invention are not limited to the objects mentioned above, and other objects not mentioned will be clearly understood by those skilled in the art from the following description.

The present invention for achieving the above object comprises the steps of (2) learning the joint angle for numerical analysis corresponding to the position of the distal end of the multi-DOF robot or the tool using the learning data; (3) calculating at least one joint angle for numerical analysis corresponding to the target position when a target position to move the distal end of the multi-DOF robot or the tool is input; and (4) calculating a target joint angle capable of reaching the distal end or tool of the multi-DOF robot to a target position from at least one joint angle for numerical analysis through inverse kinematic analysis; including, a multi-DOF robot control method is provided.

In a preferred embodiment, before the step (2), (1) a plurality of joint angles for learning and the learning arrival positions that are the positions at which the distal end of the multi-degree of freedom robot or the tool arrives according to each joint angle for learning are included. It further includes; generating the learning data.

In a preferred embodiment, the step (1) comprises the steps of: (1-1) randomly generating joint angles for learning within a range in which the joints of the multi-DOF robot can move or rotate; and (1-2) calculating the coordinate values and directions of the learning arrival position reached by the distal end of the multi-DOF robot or the tool from the joint angles for learning through regular kinematic analysis, respectively.

In a preferred embodiment, in the first step (1), after the step (1-2), (1-3) information on the movement path to move the joint of the multi-DOF robot or to limit the movement and receiving information on excitation as additional information.

In a preferred embodiment, in the first step (1), after the (1-2) step, (1-4) meeting a preset filtering condition among data including joint angles for learning and arrival positions for learning It further includes; removing the data to be used, and generating the remaining data and additional information as learning data.

In addition, the present invention provides a computer program stored in a computer readable storage medium to execute the multiple degree of freedom robot control method according to any one of claims 1 to 5 in a computer.

In addition, the present invention is a processor; a memory storing one or more instructions executable by the processor; and a communication unit for transmitting, under the control of the processor, a control signal for operating a motor or actuator provided in a joint of the multi-degree-of-freedom robot at a target joint angle to the corresponding motor or actuator, wherein the processor includes the one or more instructions By executing , the joint angle for numerical analysis corresponding to the position of the distal end of the multi-DOF robot or the tool is learned using the learning data, and when the target position to move the distal end of the multi-DOF robot or the tool is input, the target position Calculates at least one joint angle for numerical analysis corresponding to It provides a multi-degree-of-freedom robot control system.

In a preferred embodiment, the processor includes, by executing the one or more instructions, a plurality of joint angles for learning, and reaching positions for learning that are positions at which the distal end of the multiple degree of freedom or the tool arrives according to each joint angle for learning. generate training data.

In a preferred embodiment, when generating the learning data, the processor, by executing the one or more instructions, randomly generates joint angles for learning within the range in which the joints of the multi-degree of freedom robot can move or rotate, and analyze the kinematics From the joint angles for learning, the coordinate values and directions of the learning reach position reached by the tool or the distal end of the multi-DOF robot are respectively calculated.

In a preferred embodiment, when generating learning data, the processor, by executing the one or more instructions, moves the joint of the multi-DOF robot or information on the movement path to limit the movement and information on excitation induction is input as additional information.

In a preferred embodiment, when generating the learning data, the processor, by executing the one or more instructions, removes data that meets a preset filtering condition from among data including joint angles for learning and arrival positions for learning, The remaining data and additional information are generated as training data.

By means of the above-described problem solving means, the present invention learns the joint angle according to the position reached by the distal end of the multi-DOF robot or the tool, and uses the learned learning data to calculate the joint angle for numerical analysis, which is an initial solution for inverse kinematics analysis. By calculating the target joint angle that can reach the distal end or tool of the multi-DOF robot to the target position from the joint angle for numerical analysis through inverse kinematic analysis, the developer of the multi-DOF robot can determine the target corresponding to the target position. Since there is no need to directly calculate the joint angle, it is possible to reduce the burden on the developer, and also has the effect of facilitating the control of the multi-DOF robot.

In addition, in the present invention, the number of joints, shapes, and degrees of freedom of the multi-degree-of-freedom robot are limited by the burden imposed on the developer in the inverse kinematics-based calculation process. There is an effect that can solve the limitations of the number, shape, and degree of freedom of joints provided in the multi-degree-of-freedom robot.

1 is a diagram for explaining the hardware configuration of a multi-DOF robot control system according to an embodiment of the present invention.
2 is a block diagram illustrating a function performed by a processor of a multi-DOF robot control system according to an embodiment of the present invention.
3 is a block diagram for explaining a learning process for calculating joint angles for numerical analysis performed by a learning unit, which is a functional block of a processor, in a multi-DOF robot control system according to an embodiment of the present invention.
4 is a view for explaining an example of additional information input to a multi-DOF robot control system according to an embodiment of the present invention using a two-DOF robot.
5 is a view for explaining a method for controlling a multi-DOF robot according to an embodiment of the present invention.

In the following description, specific details of the invention are set forth in order to provide a general understanding of the invention, but it is common knowledge in the art that the invention may be readily practiced without these specific details and with modifications thereof. It will be self-evident to those who have

Hereinafter, a preferred embodiment according to the present invention will be described in detail with reference to the accompanying drawings 1 to 5, but mainly the parts necessary for understanding the operation and operation according to the present invention.

1 is a diagram for explaining a hardware configuration of a multi-DOF robot control system according to an embodiment of the present invention.

Referring to FIG. 1 , a multiple degree of freedom robot control system according to an embodiment of the present invention is configured to include a memory 100 , a communication unit 200 , and a processor 300 .

Here, the multiple degree of freedom robot control system according to an embodiment of the present invention is provided in the multiple degree of freedom robot and can be used to control a motor or an actuator mounted on a joint of the multiple degree of freedom robot.

In particular, the multiple-DOF robot control system according to an embodiment of the present invention moves or moves the joint of the multiple-DOF robot when a target position to move the distal end of the multiple-DOF robot or a tool mounted on the distal end is input. By calculating the target joint angle to be rotated and outputting a control signal including the calculated target joint angle to the motor or actuator, the motor or actuator mounted on the joint of the multi-DOF robot can be controlled.

In addition, the multiple degree of freedom robot control system according to an embodiment of the present invention learns the joint angle corresponding to the distal end of the multiple degree of freedom robot or the position of the tool mounted at the distal end, and uses the learning data and deep learning algorithm to learn Joint angle for numerical analysis for calculation of target joint angle, that is, an initial solution for inverse kinematics analysis can be calculated.

In addition, the multiple degree of freedom robot control system according to an embodiment of the present invention can reach the distal end of the multiple degree of freedom robot or a tool mounted on the distal end to a target position based on joint angle and inverse kinematics analysis for numerical analysis. The target joint angle can be calculated.

On the other hand, the multiple-DOF robot control system according to an embodiment of the present invention can reach the distal end or tool of the modeled multiple-DOF robot to the target position when it is desired to model the multiple-DOF robot to simulate the operation state. It can be used for calculating the target joint angle, and in addition, it can be used in various fields requiring calculation of the joint angle of the multi-DOF robot.

The memory 100 may store various data, commands, and information, in particular, programs (one or more instructions) for processing and controlling the processor 300 , and in addition, by the processor 300 It stores input or output data or learning data required for deep learning in the processor 300 .

For example, the memory 100 includes a plurality of joint angles for learning and learning data including learning arrival positions that are positions that the distal end of the multi-degree of freedom robot or tools reach according to each joint angle for learning, and operation from the learning data Information on joint angles for numerical analysis corresponding to the distal end of the multi-DOF robot or the position of the tool may be stored.

In addition, information on the target joint angle calculated from the joint angle for numerical analysis may be stored in the memory 100 .

Meanwhile, a storage medium other than the memory 100 may be separately provided, and learning data may be stored in the storage medium.

The communication unit 200 transmits a control signal including a target joint angle by communicating with a motor or actuator mounted on the joint of the multi-DOF robot. Here, the control signal including information on the target joint angle may be calculated or generated by the processor 300 .

In addition, the communication unit 200 is a UART (Universal Asynchronous Receiver Transmitter), RS-232C (Recommended Standard-232C), RS-422 (Recommended Standard-422), RS-485 (Recommended Standard-485), CAN (Controller Area Network) ), EtherCAT (Ethernet for Control Automation Technology), PLC (Power-Line Communication), PROFINET (Process Field Net), and MECHATROLINK using at least one of communication protocols including .

In addition, the communication unit 200 receives information about a target position from an external communication device, or receives information about a movement path and information on excitation guidance to move the joint of the multi-DOF robot or restrict movement. Also, various types of information for setting or controlling the processor 300 may be received.

The processor 300 may include one or more and execute one or more instructions or programs stored in the memory 100 . In addition, the processor 300 may include an AI accelerator for deep learning.

By executing one or more instructions stored in the memory 100 , the processor 300 calculates a joint angle for numerical analysis for calculation of a target joint angle using pre-built learning data.

At this time, the above-described learning data is data on the joint angle that can reach the distal end of the multi-DOF robot or the tool mounted on the distal end to a specific position, and the joint angle for numerical analysis is an initial solution for inverse kinematics analysis, It may be a position vector having a value similar to or close to the target joint angle to be calculated.

For reference, in addition to the joint angle for numerical analysis, the joint angle for learning included in the learning data and the target joint angle to be ultimately calculated are all n×1 position vectors.

In addition, when information on the target position is input, the processor 300 calculates the target joint angle by using the joint angle for numerical analysis corresponding to the target position, and at this time, based on the joint angle for numerical analysis and inverse kinematics analysis to calculate the target joint angle.

Hereinafter, a learning process for calculation of joint angles for numerical analysis performed by the processor 300 of a multi-degree-of-freedom robot control system according to an embodiment of the present invention, and a target position using joint angles for numerical analysis The calculation process of the matching target joint angle will be described in detail.

2 is a block diagram illustrating a function performed by a processor of a multiple degree of freedom robot control system according to an embodiment of the present invention, and FIG. 3 is a multiple degree of freedom robot control system according to an embodiment of the present invention. It is a block diagram for explaining a learning process for calculation of joint angles for numerical analysis performed by the learning unit 315, which is a functional block of the processor, and FIG. 4 is a multi-degree of freedom robot control system according to an embodiment of the present invention. It is a diagram for explaining an example of input additional information using the two-degree-of-freedom robot 10 .

For reference, some or all of the blocks shown in FIGS. 2 and 3 may be implemented with hardware and/or software components that perform specific functions. Functions performed by the blocks shown in FIGS. 2 and 3 may be implemented by one or more microprocessors, or circuit configurations for corresponding functions. Some or all of the blocks shown in FIGS. 2 and 3 may be software modules configured in various programming languages or script languages to be executed in the processor 300 .

2 and 3, the processor 300 of the multi-DOF robot control system according to an embodiment of the present invention includes a joint angle learning unit 310 for numerical analysis and an inverse kinematics analysis unit 320, can be configured.

The joint angle learning unit 310 for numerical analysis learns an initial solution for inverse kinematics analysis by using a deep learning algorithm, and when information on a target position (X _d ) is input, the target position (X _d ) Calculate the joint angle (q _start ) for numerical analysis as an initial solution corresponding to .

In this case, the joint angle learning unit 310 for numerical analysis may learn an initial solution using a deep neural network (DNN), and in addition, various machine learning algorithms may be applied.

In addition, the joint angle learning unit 310 for numerical analysis is a plurality of joint angles for learning (q _random ) and a position at which the distal end of the multi-degree-of-freedom robot or the tool reaches according to each joint angle for learning (q _random ). It is possible to generate learning data including the positions (X), and use this to perform learning for the calculation of the joint angle (q _start ) for numerical analysis.

As shown in FIG. 3, the joint angle learning unit 310 for numerical analysis includes a joint angle generating unit 311 for learning, a regular kinematics analysis unit 312, an additional information input unit 313, and a filter unit 314. And it may be configured to include a learning unit (315).

The joint angle generating unit 311 for learning randomly generates joint angles for learning (q _random ) within a range in which the joints of the multi-DOF robot can move or rotate. At this time, the joint angle for learning (q _random ) generated is generated as a position vector.

In addition, information about the joint angle for learning (q _random ) generated by the joint angle generating unit for learning 311 may be input to the static kinematics analyzing unit 312 and the filter unit 314 .

In addition, when generating joint angles for learning (q _random ), the range in which the joints of the multi-DOF robot can move or rotate may not be particularly limited, but in this case, multiple degrees of freedom through the filter unit 314 to be described later It is preferable to perform the process of removing the learning data out of the movable or rotatable range of the robot's joints.

The regular kinematics analysis unit 312 calculates the coordinate values and directions of the learning arrival position (X) reached by the distal end of the multi-DOF robot or the tool from the joint angles for learning (q _random ) through the regular kinematic analysis, respectively.

At this time, the regular kinematics analysis unit 312 is a Denavit-Hartenberg parameters (Denavit-Hartenberg parameters, DH parameters) from the joint angle for learning (q _random ) to calculate the coordinate value and direction of the learning arrival position (X) can

In addition, the information on the learning arrival position (X) calculated in the regular kinematics analysis unit 312 may be input to the filter unit 314 through the additional information input unit 313 .

The additional information input unit 313 receives, as additional information α, information on a movement path and excitation induction for moving or restricting movement of the joint of the multi-DOF robot through the communication unit 200 .

For reference, the additional information α input to the additional information input unit 313 is one target when the inverse kinematics analysis unit 320 calculates the target joint angle q _d corresponding to the target position X _d . If a plurality of target joint angles (q _d ) exist for the position (X _d ), it will be used as a criterion for selecting a target joint angle (q _d ) that meets the user’s needs from among the plurality of target joint angles (q _d ). can

For example, referring to FIG. 4 , a single first arm 12 connected to the base 11 , a second arm 13 and a first arm 12 connected to the first arm 12 , and A two-degree-of-freedom robot 10 provided with a joint 14 that rotatably connects the second arm 13 is shown.

In this two-degree-of-freedom robot 10, the target joint angle (q _d ) of the joint 14 to move or rotate the distal end of the second arm 13 to the target position P ₂ is P _1a and P _1b . We have two solutions containing

Here, based on the imaginary line L connecting the base 11 to the target position P ₂ , P _1a at a higher position means elbow up, and P at a relatively low position _1b means elbow down.

In addition, if information for selecting the elbow-up is input as the additional information α, the path along which the joint of the multi-DOF robot moves can be limited as the path that becomes the elbow-up. Conversely, if information for selecting elbow down as the additional information α is input, the joint of the multi-DOF robot may move in the path of elbow down.

That is, by setting such additional information α, the inverse kinematics analysis unit 320 uses the additional information α when a plurality of target joint angles q _d exist for one target position X _d . Thus, it becomes possible to select at least one target joint angle (q _d ).

On the other hand, the additional information (α) may be input in advance before calculating the coordinate values and directions of the learning arrival position (X) in the regular kinematics analysis unit 312, and the learning arrival position ( It may be input after monitoring the result value after calculating the coordinate value and direction of X).

In addition, the additional information α input to the additional information input unit 313 may be input to the filter unit 314 together with the learning arrival position X calculated from the static kinematics analysis unit 312 .

Referring back to FIG. 3 , the filter unit 314 removes data that meets a preset filtering condition from among data including joint angles for learning (q _random ) and arrival positions (X) for learning, and removes the remaining data and The additional information α is output as training data.

Here, the filtering conditions set in the filter unit 314 are, for example, that the joints of the multi-DOF robot are out of a movable or rotatable range, that the joints of the multi-DOF robot invade a limited work space, Making the joints of the multi-DOF robot move or rotate on the path where the obstacle is located, out of the range of inverse kinematic solutions set arbitrarily by the user, or inappropriate for learning for calculation of joint angle (q _start ) for numerical analysis can

For reference, the learning data in which the joints of the multi-DOF robot are out of the movable or rotatable range can be changed according to the shape or number of joints mounted in the multi-DOF robot, and the joints of the multi-DOF robot use a limited work space. The learning data that causes the multi-DOF robot to invade and the learning data that causes the joints of the multi-DOF robot to move or rotate along the path where the obstacle is located are the form or structure of the work space where the multi-DOF robot is placed and perform a specific task, and is installed in the work space. It can be changed according to other robots or structures.

In addition, learning data that is outside the range of the inverse kinematics solution set by the user and the learning data inappropriate for the calculation of the joint angle (q _start ) for numerical analysis adversely affect the joints of the multi-DOF robot or reduce the work ability. It may be learning data that degrades.

Meanwhile, except for the data filtered by the filter unit 314 , the remaining data and additional information α are generated as training data and output to the learning unit 315 .

The learning unit 315 uses the learning data to learn an initial solution corresponding to the position of the distal end of the multi-DOF robot or the tool, that is, the joint angle q _start for numerical analysis.

At this time, the learning unit 315 performs pre-processing to classify the learning data into a training dataset and an evaluation dataset, performs training and evaluation, and then calculates in the learning process. If the difference between the inverse kinematic solution and the actual solution does not exceed the threshold, learning is terminated.

In addition, if the error value during the learning evaluation is greater than the set value, the learning unit 315 may correct a hyper parameter or request additional learning data for expanding the dataset.

After that, when the target position (X _d ) to move the distal end of the multi-degree-of-freedom robot or the tool is input, the learning unit 315 receives at least one numerical value corresponding to the target position (X _d ) based on the learned data. The joint angle for analysis (q _start ) can be calculated.

Meanwhile, the joint angle q _start , the target position X _d , and the additional information α for numerical analysis calculated by the learning unit 315 may be output to the inverse kinematics analysis unit 320 .

The inverse kinematics analysis unit 320 is a target joint angle that can reach the distal end or tool of the multi-DOF robot from at least one joint angle for numerical analysis (q _start ) to the target position (X _d ) through inverse kinematics analysis. Calculate (q _d ).

In this case, the inverse kinematics analysis unit 320 may calculate the target joint angle q _d using Equation 1 below according to the inverse kinematics analysis.

[Equation 1]

q _k+1 = q _k + J ^-1 (q _k )(X _d - f(q _k ))

Here, q _k means the kth inverse kinematic solution, and the joint angle (q _start ) for numerical analysis is applied as the initial solution for this, and J is the robot Jacobian, J(q)=(∂f( q))/∂q, and X _d means the coordinate value and direction of the target position.

For reference, the inverse kinematics analysis unit 320 must iteratively update q _k in Equation 1 to calculate an inverse kinematic solution corresponding to the coordinate value and direction of the target position (X _d ), and the matrix for the robot Jacobian Matrix Inversion has to be performed, so there is a lot of calculation, and if you set a value that is too different from the target joint angle (q _d ) to be calculated and the initial solution, the calculation of the target joint angle (q _d ) is impossible or requires a lot of calculation time. may be consumed

However, the inverse kinematics analysis unit 320 uses the joint angle for numerical analysis (q _start ) calculated in the learning unit 315 as an initial solution, and the aforementioned joint angle for numerical analysis (q _start ) is the target to be calculated. Since the joint angle (q _d ) is a position vector having the same or approximate value, the amount of calculation and the calculation time of the inverse kinematics analysis unit 320 can be significantly reduced, and the calculation of the target joint angle (q _d ) is impossible. can be prevented in advance.

In addition, since the developer of the multi-DOF robot does not need to directly calculate the target joint angle (q _d ) corresponding to the target position (X _d ), it is possible to reduce the developer’s burden, and There is no need to limit the number, shape, or degree of freedom.

So far, a multi-DOF robot control system according to a preferred embodiment of the present invention has been described. Hereinafter, a method for controlling a multi-DOF robot according to a preferred embodiment of the present invention will be described. However, since the multiple-DOF robot control method according to the preferred embodiment of the present invention is performed in the above-described multiple-DOF robot control system, the functions performed in each step are as described above in relation to the above-described multiple-DOF robot control system. Same as function. Therefore, in the following, the overall flow will be mainly described.

5 is a diagram for explaining a method for controlling a multi-DOF robot according to an embodiment of the present invention.

Referring to FIG. 5 , first, the processor of the multiple degree of freedom robot control system according to an embodiment of the present invention provides a plurality of joint angles for learning and the distal end or tool of the multiple degree of freedom robot reaches according to each joint angle for learning. The training data including the training arrival positions that are positions is generated (S100).

Specifically, the processor randomly generates joint angles for learning within the range in which the joints of the multi-DOF robot can move or rotate (S110), and the distal end or tool of the multi-DOF robot from the joint angles for learning through regular kinematic analysis. Coordinate values and directions of the learning arrival positions to be reached are respectively calculated (S120).

At this time, the created joint angle for learning is created as a position vector, and the coordinate value and direction of the learning destination can be calculated from the joint angle for learning using Denavit-Hartenberg parameters (DH parameters) in the regular kinematics analysis. can

Then, the processor receives the information on the movement path to move the joint of the multi-DOF robot or restrict the movement and information on the excitation as additional information (S130), and includes joint angles for learning and arrival positions for learning. After removing data that meets a preset filtering condition from among the collected data, the remaining data and additional information are generated and output as learning data (S140).

In this case, the above-described additional information includes, when a target joint angle corresponding to the target position is calculated, when a plurality of target joint angles exist for one target position, a target joint angle that meets the user's request among the plurality of target joint angles can be used as a criterion for selecting

In addition, the preset filtering conditions are, for example, that the joints of the multi-DOF robot are out of a movable or rotatable range, that the joints of the multi-DOF robot invade a limited work space, and the joints of the multi-DOF robot are It may be that they move or rotate in the path where the obstacle is located, that it is outside the range of the inverse kinematic solution set arbitrarily by the user, or that it is not suitable for learning for the calculation of joint angles for numerical analysis.

Next, the processor learns the joint angle for numerical analysis corresponding to the position of the distal end of the multi-DOF robot or the tool by using the learning data (S200).

At this time, the processor performs pre-processing on the training data, divides it into a training dataset and an evaluation dataset, performs training and learning evaluation, and then performs the inverse kinematic solution calculated in the learning process and the actual If the difference in solutions does not exceed the threshold, learning may be terminated.

In addition, if the error value is greater than the set value in the learning evaluation process, a hyper parameter may be corrected or additional learning data may be requested for data set expansion.

Then, when the target position to move the distal end of the multi-DOF robot or the tool is input, the processor calculates at least one joint angle for numerical analysis corresponding to the target position (S300).

Then, the processor calculates a target joint angle capable of reaching the distal end of the multi-DOF robot or the tool to a target position from at least one joint angle for numerical analysis through inverse kinematics analysis (S400).

At this time, the processor performs inverse kinematics analysis by using the joint angle for numerical analysis as an initial solution. Since the joint angle for numerical analysis is a position vector having the same value or close to the target joint angle to be calculated, inverse kinematics analysis The amount of calculation and calculation time involved in the process can be remarkably reduced, and the situation in which the calculation of the target joint angle is impossible can be prevented in advance.

For this reason, since the developer of the multi-DOF robot does not need to directly calculate the target joint angle corresponding to the target position, the burden on the developer can be reduced, and the number, shape, and degree of freedom provided in the multi-DOF robot can be limited. there will be no need

For reference, in the step S400, the example of calculating the target joint angle that can reach the distal end of the multi-DOF robot or the tool to the target position from the joint angle for numerical analysis has been described above with reference to Equation 1 , and detailed descriptions are omitted.

The present invention as described above can also be implemented as computer-readable codes in a computer-readable storage medium. The computer readable storage medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device. In addition, the computer-readable storage medium may be distributed in network-connected computer systems, and the computer-readable code may be stored and executed in a distributed manner.

So far, the present invention has been looked at with respect to preferred embodiments thereof.

Those of ordinary skill in the art to which the present invention pertains will understand that the present invention can be implemented in a modified form without departing from the essential characteristics of the present invention. Therefore, the disclosed embodiments are to be considered in an illustrative rather than a restrictive sense. The scope of the present invention is indicated in the claims rather than the foregoing description, and all differences within the scope equivalent thereto should be construed as being included in the present invention.

100 : memory
200: communication department
300 : processor

Claims (11)

(2) learning the joint angle for numerical analysis corresponding to the position of the distal end of the multi-DOF robot or the tool by using the learning data;
(3) calculating at least one joint angle for numerical analysis corresponding to the target position when a target position to move the distal end of the multi-DOF robot or the tool is input; and
(4) calculating a target joint angle capable of reaching the distal end of the multi-DOF robot or the tool to a target position from at least one joint angle for numerical analysis through inverse kinematics analysis; including, controlling the multi-DOF robot method.
The method of claim 1,
Before step (2),
(1) generating learning data including a plurality of joint angles for learning and learning arrival positions that are positions reached by a distal end or a tool of the multi-DOF robot according to each joint angle for learning; characterized by further comprising: A method for controlling a multi-DOF robot.
3. The method of claim 2,
The first step (1) is,
(1-1) randomly generating joint angles for learning within a range in which the joints of the multi-DOF robot can move or rotate; and
(1-2) calculating the coordinate values and directions of the learning arrival position reached by the distal end of the multi-DOF robot or the tool from the joint angles for learning through regular kinematic analysis, respectively; Road robot control method.
4. The method of claim 3,
The first step (1) is, after the step (1-2),
(1-3) receiving, as additional information, information on a movement path and information on excitation guidance to move the joint of the multi-degree-of-freedom robot or to restrict movement; Road robot control method.
4. The method of claim 3,
The first step (1) is, after the step (1-2),
(1-4) removing data that meets a preset filtering condition from among data including joint angles for learning and arrival positions for learning, and generating the remaining data and additional information as learning data; characterized by further comprising: A method for controlling a robot with multiple degrees of freedom.
A computer program stored in a computer-readable storage medium to execute the method for controlling the multi-DOF robot according to any one of claims 1 to 5 in a computer.
processor;
a memory storing one or more instructions executable by the processor; and
A communication unit for transmitting a control signal for operating a motor or actuator provided in a joint of the multi-DOF robot at a target joint angle to the corresponding motor or actuator under the control of the processor; includes;
The processor, by executing the one or more instructions,
Using the learning data, learn the joint angle for numerical analysis corresponding to the position of the distal end of the multi-degree-of-freedom robot or the tool,
When the target position to move the distal end of the multi-DOF robot or the tool is input, at least one joint angle for numerical analysis corresponding to the target position is calculated,
A multi-DOF robot control system that calculates a target joint angle that can reach the distal end or tool of the multi-DOF robot to a target position from at least one joint angle for numerical analysis through inverse kinematic analysis.
8. The method of claim 7,
The processor, by executing the one or more instructions,
A multiple degree of freedom robot control system for generating learning data including a plurality of joint angles for learning, and arrival positions for learning that are positions reached by a distal end or a tool of a multiple degree of freedom robot according to each joint angle for learning.
9. The method of claim 8,
When generating training data, the processor, by executing the one or more instructions,
The joint angles for learning are randomly generated within the range where the joints of the multi-degree-of-freedom robot can move or rotate,
A multi-degree-of-freedom robot control system that calculates the coordinate values and directions of the learning reach position reached by the distal end of the multi-degree-of-freedom robot or the tool from the joint angles for learning through regular kinematic analysis, respectively.
10. The method of claim 9,
When generating training data, the processor, by executing the one or more instructions,
A multi-degree-of-freedom robot control system that receives information on the movement path and excitation guidance to move the joint of the multi-DOF robot or restrict movement as additional information.
10. The method of claim 9,
When generating training data, the processor, by executing the one or more instructions,
A multiple degree of freedom robot control system that removes data that meets a preset filtering condition from among data including joint angles for learning and arrival positions for learning, and generates the remaining data and additional information as learning data.

Images