JP2008030167A - Motion acquisition system and its control method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To establish a stable contact operation with an unknown actual environment, and to extract, conserve and reproduce each motion like fingers of a person. <P>SOLUTION: An acceleration control for making control rigidity zero while keeping robustness is performed for an actuator 1 of each robot system by a disturbance observer. The disturbance acceleration θ''<SP>ext</SP>and the acceleration response θ''<SP>res</SP>of the actuator 1 are converted into acceleration disturbance m''<SP>ext</SP>and an acceleration response value m''<SP>res</SP>, respectively, in a non-interference mode of an orthogonal coordinate system by using a mode quarry matrix by mode acceleration conversion means 27 and 28. A motion in relation to an object 47 corresponding to each of non-interference modes is extracted based on the acceleration disturbance m''<SP>ext</SP>and the acceleration response value m''<SP>res</SP>. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a motion learning system that enables extraction, storage, and reproduction of motion of, for example, an expert, and a control method thereof.

  In recent years, there has been a demand for establishment of techniques for supporting human advanced work such as minimally invasive surgical treatment by robots and skill preservation of skilled workers. In response to these demands, many studies have been conducted to extract human skills based on image information, but the extraction and reproduction of tasks that repeat contact with the environment (human work content) is a force control. This is a difficult situation because an operation based on the above is required.

For example, in Patent Document 1, when the master mechanism recognizes writing information and work information related to a writing operation on the inner surface of a plate as a workpiece, the controller controls the operation of the robot arm of the slave mechanism and the slave side plate, A system for reproducing the writing operation is disclosed. However, as with the image information, since the motion is extracted and reproduced only by the position information, it is difficult to reproduce the repeated operation of contact with or contact with the environment.
JP 2006-62052 A

  Other techniques for resolving these problems have been proposed, such as reproduction in a virtual environment, or using hybrid position and force control. However, at present, a satisfactory force control system that reproduces according to human actions and task contents has not been instrumented, and implementation is not always easy. The reason for this is that although force information is originally bidirectional information that is governed by the “law of action / reaction”, the current force control system is unidirectional and bilateral (Bilateral). This is thought to be due to the lack of basic methods for handling sexual information. Also, in order to realize robotic dexterous movements like humans, an efficient integrated motion of a multi-degree-of-freedom system has been strongly desired.

  The present invention has been made in view of the above-mentioned problems, and its purpose is to establish a stable contact operation with an unknown real environment and to extract a motion such as a human finger individually. It is an object of the present invention to provide a learning system control method and apparatus.

  Another object of the present invention is to provide a control method and apparatus for a motion learning system capable of storing the extracted motion.

  Furthermore, another object of the present invention is to provide a control method and apparatus for a motion learning system capable of realizing efficient integrated motion of a multi-degree-of-freedom system and reproducing position-force hybrid motion. There is.

  In order to solve the above-mentioned problem, a first step of performing acceleration control to zero control rigidity while maintaining robustness on each actuator of a plurality of robot systems arranged corresponding to the degree of freedom of operation in real space; The equivalent acceleration of external force to each actuator and the acceleration response of each actuator are converted into the equivalent acceleration and acceleration response of external force in independent non-interfering mode defined by orthogonal coordinate system using mode queuing matrix And a third step of extracting a motion for the object corresponding to each of the non-interference modes based on the equivalent acceleration and acceleration response of the external force in the non-interference mode obtained in the second step. Steps.

  In the control method, the robot system includes a master system and a slave system, and the actuators of the master system and the slave system are controlled by bilateral control based on the acceleration control in the first step. It is preferable.

  Moreover, it is preferable to further include a fourth step of storing the acceleration information converted into the non-interference mode.

  Further, for each non-interference mode, a ratio of an acceleration response to an equivalent acceleration of external force in the non-interference mode is determined, an acceleration target value based on the acceleration information stored in the fourth step, and the second step A fifth step of calculating the acceleration reference value from the acceleration control system in the non-interference mode set based on the ratio, using the deviation from the equivalent acceleration or acceleration response of the external force in the non-interference mode obtained in Preferably, the method further includes a sixth step of converting an acceleration reference value in the non-interference mode into an acceleration reference value for each robot system using an inverse matrix of a mode quality matrix.

  The actuator is preferably a linear actuator.

  The motion acquisition system according to the present invention includes an acceleration control unit that performs acceleration control for making the control rigidity zero while maintaining robustness for each actuator of a plurality of robot systems arranged corresponding to the degree of freedom of operation in real space, A mode acceleration that converts the equivalent acceleration of external force to each actuator and the acceleration response of each actuator into an equivalent acceleration and acceleration response in the independent non-interference mode defined in the Cartesian coordinate system using a mode-quarty matrix. A conversion means; and a motion extraction means for extracting a motion for the object corresponding to each of the non-interference modes based on the equivalent acceleration and acceleration response in the non-interference mode obtained by the mode acceleration conversion means. Yes.

  In the above system, the robot system includes a master system and a slave system, and the acceleration control means controls the actuators of the master system and the slave system by bilateral control based on the acceleration control. Preferably there is.

  Moreover, it is preferable to further comprise storage means for storing acceleration information converted into the non-interference mode.

  Further, for each non-interference mode, the ratio of the acceleration response to the equivalent acceleration of the external force in the non-interference mode is determined, the acceleration target value based on the acceleration information stored in the storage means, and the mode acceleration conversion means Mode acceleration calculation means for calculating an acceleration reference value from an acceleration control system in the non-interference mode set based on the ratio, using the obtained deviation from the equivalent acceleration or acceleration response of the external force in the non-interference mode as an input; and the mode It is preferable that the apparatus further includes a mode acceleration reverse conversion unit that converts an acceleration reference value in the non-interference mode into an acceleration reference value for each robot system using an inverse matrix of a quarry matrix.

  The actuator is preferably a linear actuator.

  According to the method of claim 1 and the system of claim 6, acceleration control is performed for each actuator of a plurality of robot systems so that the control rigidity is zero without losing robustness. The operation involving the can be realized. In addition, it is possible to newly define a non-interfering mode of the Cartesian coordinate system based on human movements and tasks using the mode queuing matrix, and each can extract independent motions. Extraction of motion corresponding to is easy. Therefore, a stable contact operation with an unknown real environment can be established, and motions such as human fingers can be extracted individually.

  According to the method of claim 2 and the system of claim 7, it is possible to extract a reaction force from an environment particularly important for motion extraction.

  According to the method of claim 3 and the system of claim 8, it is possible to store the extracted motion and read acceleration information converted into, for example, the non-interference mode at another location.

  According to the method of claim 4 and the system of claim 9, the motion information can be reproduced by the same or different robot system based on the acceleration information converted into the non-interference mode extracted and stored previously. Sex is secured. In particular, since the control system is configured in the acceleration dimension for each non-interference mode, position information and force information can be integrated, and the ratio of the acceleration response to the equivalent acceleration of external force in the non-interference mode is 1 or 0. If it is set to, reproduction in the position dimension or the force dimension becomes possible for each non-interference mode, and this can be reflected in the motion of the robot system in the real space. Therefore, efficient integrated motion of a multi-degree-of-freedom system can be realized and hybrid motion of position and force can be reproduced.

  According to the method of claim 5 and the system of claim 10, the influence of friction can be mechanically removed by the linear actuator that performs linear driving.

  Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

  1 to 10 show a first embodiment of the present invention. First, an example of a robot system that realizes acceleration control in motion control, which can be applied to the motion learning system of the present invention, will be described with reference to FIG.

In the figure, reference numeral 1 denotes a movable actuator that is provided as a target for estimating disturbance torque and converts energy into motive power. This is the actual position by inputting a current value i to a drive source such as a linear motor. It operates at θ _m . The actuator 1, in order to detect converts the actual location theta _m into an electrical position signal, for example, the position sensor 2 such as a linear encoder is mounted.

On the other hand, a position information type disturbance observer 11 for controlling the actuator 1 inputs a current value i which is a command value to the actuator 1 and a value obtained by pseudo-differentiating the position θ _m by the pseudo-differentiator 12, that is, a response speed. Value ^ θ ^· _m (Hereinafter, except for mathematical formulas, “^” representing this estimated value is written before the corresponding symbol, and the first-order differentiation is “ ^· ”, and the second-order differentiation is “ ^·· ”.) For the sake of convenience, and is also written after the corresponding symbol.) Is input, and the estimated value of the disturbance torque is output from each of these values. The Pseudo differentiator 12, converts the position signals to speed signals, by performing pseudo differential to position theta _m, it is possible to obtain a conversion output which suppresses the sensitivity to noise. The disturbance observer 11 incorporates an inverse model equivalent to the actuator 1 and converts the current i into a first signal in units of torque (force). The first signal and the response speed ^ from the pseudo-differentiator 12 an inverse model unit 14 for outputting a third signal obtained by comparing the second signal obtained by differentiating θ _m ^· , and setting a cutoff frequency upon differentiation in the inverse model unit 14; A low-pass filter 15 that extracts a third signal having a low frequency band component from the inverse model unit 14 and outputs the third signal as an estimated value ττ ^dis of the disturbance torque. Reference ^numeral 16 denotes torque-current conversion means for inversely converting the estimated value τ ^dis of the disturbance torque into an estimated current value I ^cmp in the same unit as the target current reference value I ^ref , and this estimated current value I ^cmp And the current reference value I ^ref are added by an adder 17 to calculate the current value (current command value) i input to the actuator 1. Thereby, in the apparatus having each configuration of FIG. 1, the disturbance torque τ ^dis applied to the actuator 1 from the outside can be removed. Since the actuator 1 outputs a position signal, the output of the actuator 1 (acceleration θ _m ^... Is integrated twice to output the actual position θ _m through a two-stage integrator (1 / s) for convenience. ).

The robot system shown in FIG. 1 constructs a robust acceleration control system by removing the influence of various disturbances acting on the motor by the disturbance observer 11. Disturbance torque τ ^dis acting on the motor is expressed by the following ^{equation (} 1).

Here, each symbol described in FIG. 1 and Equation 1 will be described. I ^ref is a current reference value, J _m is a motor inertia, K _t is a torque constant, and θ _m is a motor position. And the subscript _n after the symbol means a nominal value. Further, the first term on the right side of Equation 1 is the inertia fluctuation torque, the second term is the torque ripple due to the fluctuation of the torque constant, the third term F _c is the Coulomb friction torque, and the fourth term D _m θ _m ^· is the viscous friction torque. Τ ^{ext in} the fifth term represents an external force.

When the current reference value I ^ref and the motor speed θ _m ^· can be detected by the pseudo-differentiator 12 having the cutoff frequency g _pd , the disturbance torque τ ^dis defined by the equation 1 is the disturbance observer 11 shown in FIG. Is estimated through the first-order low-pass filter 15 as follows.

However, in the above equation 2, g * _dis can be expressed as follows.

By converting the estimated value τ ^dis of the disturbance torque into the estimated current value I ^cmp by the torque-current conversion means 16 and feeding back, a robot system having a control system robust to the disturbance can be constructed. .

FIG. 2 is an equivalent representation of the robot system shown in FIG. The robot system in FIG. 1 is affected by the sensitivity function of the mathematical formula shown in block 21 of FIG. That is, the disturbance observer 11 can estimate only the external force applied to the robot system. Note 22, the difference between the disturbance torque estimated acceleration reference value theta _m ^{· · ref} is an equivalent representation of the subtracter to be outputted as an acceleration response value theta _m ^{· ·.}

As shown in FIG. 2, it can be seen that the robust control system based on the disturbance observer 11 is an acceleration control system that controls the motor of the actuator 1 in the acceleration dimension. In FIG. 2, it can be seen that the influence of the disturbance torque τ ^dis is eliminated by setting the cutoff frequency g _dis of the disturbance observer 11 to be large. Thereby, a robust acceleration control system for the motor constituting the actuator 1 can be realized.

  Here, the reason why the acceleration control system robot system is employed in the motion extraction, storage, and reproduction (collectively, “learning”) system of the present invention will be described. With respect to the control stiffness of the system, which is a design index for motion control, when the driving force of the actuator 1 is f and the position is x, the control stiffness κ can be defined as follows.

According to Equation 4, the relationship between position control and force control can be understood as follows.

In ideal position control, since the displacement of the position x becomes 0, the control rigidity κ becomes infinite, and any amount can be used if the force f is a finite value, that is, the force f becomes indefinite. On the other hand, in ideal force control, the displacement of the force f is zero, so the control rigidity κ is zero and the position x is indefinite.

  In motion control for human interaction, soft control with a control rigidity κ of 0 is desired. In order to reduce the control rigidity κ to 0 without losing the robustness, a method for removing the equivalent disturbance determined by the disturbance observer 11 by removing the position control loop with respect to the actuator 1 is widely known. As a result, the robot system employed in the present invention inevitably becomes an acceleration control system that realizes a desired acceleration.

  Next, a method of motion extraction using the acceleration control system robot system will be described. Here, interaction mode control is introduced as an integrated control method for a robot system group for extracting human multi-degree-of-freedom motion. In this interaction mode control, unlike the conventional method of configuring the control system for each individual system, only the interference between the robot systems is considered, and the control system is configured in a non-interfering mode space. The modes between the robot systems can be extracted by introducing a mode-queuing matrix to be described later.

  In addition, a new control index called a hybrid ratio is introduced as a task code for describing tasks to be achieved by the robot group. In the interaction mode control, all information is mutually exchanged on an acceleration basis. Therefore, an intuitive and flexible distributed system design is possible by arbitrarily determining a hybrid ratio in each mode space.

The concept of the hybrid ratio will be explained. First, when the robustness of the robot system is ensured by the disturbance observer 11 as shown in FIG. 1, the system is equivalently converted into an acceleration control system as shown in FIG. Is done. In Figure 3, the input to the robot system is output as acceleration disturbance theta ^{.. ext} next, ^... the deviation subtracter 22 acceleration response value theta _m from another robot system and the acceleration reference value theta _m ^{... ref} Is equivalently configured. In the ideal position control, the acceleration reference value θ _m ^{·· ref} and the acceleration response value θ _m ^·· can be controlled to be equal to each other in order to completely suppress the influence of the external force. On the other hand, in the ideal force feedback control (force control), since the external force is a command value input to the system, the acceleration disturbance θ ^·· _ext and the acceleration response value θ _m ^·· are equal. Here, the hybrid ratio HR introduced as an index of control system design in the acceleration control system is defined as the ratio of the influence of the acceleration disturbance θ ^·· _ext by the other system on the acceleration response value θ _m ^·· as follows: Is done.

According to the above formula, the hybrid ratio HR is 0 in ideal position control, and the hybrid ratio HR is 1 in ideal force feedback control. The relationship between the proposed hybrid ratio HR and the control rigidity κ is as shown in Table 1 below.

Next, the mode quality matrix will be described. Since the interaction mode control controls only the interference between a plurality of robot systems, this embodiment uses a mode quality matrix in the orthogonal coordinate system. Extract the defined non-interference mode. By using a mode-queuing matrix that is a parameter conversion means, the physical parameters of the robot system can be converted into parameters in a virtual mode space processed in the computer. The mode quality matrix Q _n (n = 1 to 5) is defined as follows, for example.

The degree of the mode queuing matrix Q _n matches the number of task degrees of freedom that can be realized by the robot system group. For this reason, for example, when a number of robot systems corresponding to the motion of a human finger are prepared, the mode quality matrix Q _n is determined with an order that matches the number of robot systems. Further, as the nature of the mode qualifiers over matrix Q _n, a product of the mode qualifiers over matrix Q _n and its transposed matrix to become orthogonal matrix it can be confirmed to have orthogonality. Thus, instead of designing a control system for each individual robot, the mode qualifiers over matrix Q _n designed non-interference has been a control system in a virtual orthogonal mode on, the inverse matrix of the mode qualifiers over matrix Q _n A certain inverse mode quality matrix Q _n ⁻¹ makes it possible to integrate the acceleration reference values θ _m ^{·· ref} in each robot system configured with multiple degrees of freedom.

FIG. 4 is a block diagram showing a general structure of the interaction mode control. Here, the distributed control design of the robot system group by the interaction mode control realized as the motion learning system is shown. In the figure, θ ^·· indicates the acceleration in the joint coordinate system of the robot system. In the joint space, the first subscripts A to N represent the identifiers of the robot systems in alphabetical order, and the second subscripts 1 to N represent the identifiers of the actuators 1 in the respective robot systems in numerical order. . Further, the subscripts A to N in the work space are assigned corresponding to the identifiers of the robot systems.

In FIG. 4, N robot systems that perform general joint motion are each provided with one to a plurality of actuators 1 and arranged in a joint space that is a real space. Although not shown here, each actuator 1 includes a disturbance observer 11 including a pseudo-differentiator 12 and a torque-current conversion means 16 in order to realize a robust acceleration control system as shown in FIG. Is provided. The acceleration disturbance theta ^{· · ext} from ^{· · res} and another robot system acceleration response values theta for each actuator 1 conversion, by the acceleration conversion means 25 and 26, the acceleration x ^{· ·} in the work space using the Jacobian matrix It can be done. Here, the first acceleration converting means 25 takes in the acceleration disturbance θ ^{·· ext} to each actuator 1 constituting one robot system and converts it into the acceleration disturbance x ^{·· ext} in the work space. Another second acceleration conversion means 26 takes in the acceleration response values θ ^{·· res} from the actuators 1 constituting one robot system and converts them into acceleration response values x ^{·· res} in the work space. Preferably, the acceleration disturbance θ ^{·· ext} is estimated and obtained by a disturbance observer 11 including a pseudo-differentiator 12, thereby realizing a force sensorless in each actuator 1 and simplifying the system configuration. . The acceleration response value θ ^{·· res} from the actuator 1 can also be obtained by differentiating the detection signal from the position sensor 2 described above twice with a differentiator (not shown).

Reference numerals 27 and 28 _denote mode acceleration conversion means for storing and storing the above-described mode quality matrix Qn. The mode acceleration conversion means 27, 28 receives the acceleration response values x ^{·· ref} obtained by the acceleration conversion means 25, 26, extracts the interference of the robot system group in the work space, and has a built-in mode quarry matrix by using Q _n, it is to be converted into non-interfering acceleration m ^{· ·} output in the orthogonal mode space. Specifically, the first mode acceleration conversion means 27 converts the acceleration disturbance x ^{·· ext} in the work space obtained by the first acceleration conversion means 25 into the acceleration disturbance m ^{·· ext} in the mode space. The second mode acceleration conversion means 28 converts the acceleration response value x ^{·· res} in the work space obtained by the second acceleration conversion means 26 into the acceleration response value m ^{·· res} in the mode space. The acceleration disturbance m ^{·· ext} and the acceleration response value m ^{·· res} obtained in this way are output to the interaction mode control unit 31 described later as acceleration information of a plurality of independent modes corresponding to the number of robot systems. And preferably stored in the storage means 30 of the computer.

31 is an input of acceleration disturbance m ^{·· ext} , acceleration response value m ^{·· res} , acceleration target value m ^{·· cmd} and task code in the mode space, and motion extraction in the mode space and reproduction of motion by the robot system The interaction mode control unit generates the acceleration reference value m ^{·· ref} in each mode space. Reference numeral 32 denotes an individual mode space in which the interaction mode control unit 31 sets a desired hybrid ratio HR that defines the ratio between the equivalent acceleration of the external force and the acceleration response in response to a task command set in the computer. It is a hybrid ratio determining means that determines each time and outputs the result as a task code to the interaction mode control unit 31. When the hybrid ratio determining unit 32 receives a task command indicating that an “extraction mode” to be described later is selected by an operating unit (not shown), for example, the hybrid ratio determining unit 32 determines a hybrid ratio considering the configuration of the actual robot system, When a task command indicating that “mode” has been selected is received, a hybrid ratio corresponding to a desired mode space defined by the mode-queuing matrix Q _n is determined, and a task code corresponding to any one of the hybrid ratios is determined. It is designed to generate.

Further, 33 includes an inverse mode quality matrix Q _n ⁻¹ for inversely transforming the acceleration reference value m ^{·· ref} of each mode obtained by the interaction mode control unit 31 into an acceleration reference value x ^{·· ref} in the work space. Mode acceleration reverse conversion means. The acceleration reference value x ^{·· ref} in the work space obtained by the mode acceleration inverse conversion means 33 is referred to the acceleration reference value in the joint space by the acceleration inverse conversion means 34 having an inverse Jacobian matrix provided corresponding to the number of robot systems. It is converted back to the value θ ^{·· ref} and output to each actuator 1. Thereby, based on the acceleration reference value m ^{·· ref} of each mode output from the interaction mode control unit 31, it is possible to reproduce the motion in each robot system arranged in the real space.

  Note that all the components except the actuator 1 are provided as computer processing means. Therefore, each block configuration shown in FIG. 4 is only functionally partitioned for easy understanding, and any processing unit having an equivalent function is included in the scope of the present invention.

In each configuration described above, the acceleration m ^·· of the mode space converted by the mode acceleration conversion means 27 and 28 is expressed in the orthogonal coordinate system as the property of the mode quarantine matrix Q _n , and therefore independent in each mode space. It is possible to design a simple control system. In other words, since the hybrid ratio HR corresponding to the desired task can be realized for each mode space, the hybrid ratio HR of each mode can be associated as a task code by the hybrid ratio determining means 32. The acceleration reference value m ^{·· ref} generated in each mode space is obtained as an acceleration reference value θ ^{·· ref} of each robot system by an inverse mode queuing matrix Q _n ^-1 prepared in the mode acceleration inverse conversion means 33. Integrated and realized. Thus, the interaction mode control by interaction mode control unit 31 here is similar to the kinematics of the robot using Jacobian matrix, as a new orthogonal mode coordinate conversion method according to mode qualifiers over matrix Q _n, motion of the robot system group Integration can be realized.

Next, each method for extracting and reproducing a human object gripping motion based on the above interaction mode control will be described. When the operation object is a human finger, the order of the mode queuing matrix Q _n corresponds to the number of human fingers that can be reproduced, so that human skilld motion can be easily acquired and analyzed in an independent mode space. Can be performed. Here, a series of operations of “extraction mode” for extracting and storing human motion information in the mode space and “reproduction mode” for reproducing motion based on the storage information and task code are described. explain.

  FIG. 5 is an example of an experimental apparatus for realizing the apparatus configuration shown in FIG. The experimental device here applies the interaction mode control to the three linear motor systems. Therefore, it is possible to extract and reproduce motion corresponding to the three fingers of the human thumb, index finger, and middle finger. In the experimental apparatus of FIG. 5, a rod-type linear actuator capable of mechanically removing the influence of friction is adopted for each of the three robot systems A, B, and C arranged in the real space. This is arranged as a linear motor 44 having a rod-shaped movable portion 43 that is movable in the horizontal direction with respect to the base 42. Reference numeral 46 denotes a linear encoder corresponding to the position sensor 2. Detailed parameters of each actuator were set as shown in Table 2 below. The sampling time in the experiment is set to 100 μs.

In FIG. 5, reference numeral 47 denotes an object supported at the tip of the movable part 43 of each robot system A, B, C. Here, the robot systems A, B, and C are regarded as three fingers, and the object 47 is grasped from both sides by the movable parts 43 of the robot system A and the movable parts 43 of the robot systems B and C. .

Since the system configured in this way has three degrees of freedom, the acceleration response values x ^{·· ref} and acceleration disturbance x ^{·· ext} of each actuator are the third-order mode ^qualities of the mode acceleration conversion means 27 and 28, respectively. The matrix Q ₃ can be used to convert acceleration response values m ^{·· res} and acceleration disturbances m ^{·· ext} on the mode coordinates shown in the following equations (8) and (9).

^{_{Here, x i ·· (i = A~C}} ) shows the acceleration in each of the robotic system ^{_{A~C, m i ·· (i =}} 1~3) represents acceleration in the modal space. As shown in FIG. 6, using the mode qualifier chromatography matrix _{Q 3,} the acceleration _{x i} ^{· · res} in the robot system A through C, _{x i} a ^{· · ext,} acceleration _{m i} ^{· · res} in the respective mode space, m _By converting to _i ^{·· ext} , the center of gravity (COG) as the first mode, the yawing as the second mode, and the three independent modes inside as the third mode can be converted. The reason for the conversion to such a mode is that when three human fingers are considered, the motion is an actuator that moves the center of gravity of the target object 47, an actuator that rotates the target object 47, and an actuator that grips the target object 47. It can be decomposed into three independent modes. When the operation article is an object gripping task, each mode corresponds to a movement task, a rotation task, and a gripping task. Therefore, virtual actuators (tasks) defined in the mode space may be set other than the above-mentioned COG, yawing, and internal operation according to the operation object to be considered. There may be two or more independent modes.

  Switching between the extraction mode and the reproduction mode can be realized by a task code integrated with a hybrid ratio. That is, when any one of the extraction mode and the reproduction mode is instructed by a task command, the hybrid ratio determining unit 32 outputs a hybrid ratio having a different value for each mode to the interaction mode control unit 31 as an integrated task code. It is like that. The task codes of the three robot systems A to C can be defined as follows.

FIG. 7 shows a block diagram of a motion learning system that realizes extraction and reproduction control of human object gripping motion based on interaction mode control, and the same parts as those in FIG. 4 are denoted by the same reference numerals. . Since the experimental apparatus used here is a one-degree-of-freedom system using a linear motor as a drive source for the robot systems A to C, the work space shown in FIG. 4 and the joint space are the same. Therefore, in such a system, the acceleration conversion means 25 and 26 and the reverse acceleration conversion means 34 as shown in FIG. 4 are unnecessary, and the acceleration disturbance x ^{·· ext} of each robot system A to C installed in the joint space is The acceleration response values x ^{·· res of the} robot systems A to C are directly input to the second mode acceleration conversion means 28. Further, the acceleration reference value m ^{·· ref} generated in the mode space is converted into the acceleration reference value x ^{·· ref} in the joint space by the mode acceleration inverse conversion means 33 using the inverse mode quality matrix Q ₃ ^-1. , Distributed and integrated to the actuators 1 of the robot systems A to C.

Further, in FIG. 7, reference numeral 51 denotes an acceleration reference value m ^{·· ref} or acceleration of the mode space converted by the mode acceleration conversion means 27 and 28 in accordance with the hybrid ratio determined by the hybrid ratio determination means 32 and the task code. This is task code selection means for sending one of the disturbances m ^{·· ext} to the interaction mode control unit 31 for each unit of the mode space. The interaction mode control unit 31 includes subtractors 52A to 52C that subtract one of the acceleration reference value m ^{·· ref} or acceleration disturbance m ^{·· ext} from the acceleration target value m ^{·· cmd for} each mode space, The acceleration reference value m ^{·· ref} in the mode space is calculated by multiplying the calculation results of the subtractors 52A to 52C obtained for each mode space by the built-in transfer elements C ₁ (s) to C ₃ (s). Integrating means 53A to 53C.

  In the above configuration, in the extraction mode, the three robot systems A to C operate in accordance with human operation input, so it is necessary to control the reaction force from the human. Therefore, when a task command indicating that the extraction mode has been selected is output to the hybrid ratio determining unit 32, the hybrid ratio determining unit 32 sets the respective hybrid ratios HR to ideal force control in the three non-interference modes described above. Set to 1. The task code at this time is as follows.

Further, in order to realize the task code represented by Equation 11, each transfer element C _i (s) (i = 1 to 3) of the control system in the mode space is configured as a force servo as follows. Here, K _fi (i = 1 to 3) is a force servo gain.

At the same time, since the hybrid ratio HR of each mode is 1, the hybrid ratio determining means 32 is acceleration disturbance m ₁ ^{·· ext} , m _{2 of} each mode converted by the first mode acceleration converting means 27. ^.. , M ₃ ^{... The} task code selection means 51 is controlled so that ^ext is input to the interaction mode control unit 31. In the extraction mode, since the acceleration target values m ₁ ^{·· cmd} , m ₂ ^{·· cmd} , m ₃ ^{·· cmd} for each mode space are 0, the interaction mode control unit 31 determines the acceleration disturbance m in each mode. ₁ ^{·· ext} , m ₂ ^{·· ext} , m ₃ ^{·· ext} is multiplied by the gain of the force servo described in ^Equation 12 to obtain the acceleration reference value m ₁ ^{·· ref} , m ₂ ^··· in the mode space. ^ref , m ₃ .. ^ref is calculated. The acceleration reference value ^{_{m 1 ·· ref, m 2 ··}} ref, m 3 ·· ref is the reverse mode qualifiers over matrix _Q ^{3 -1} in the mode acceleration inverse transformation means 33, an acceleration reference value ^{x · ·} in the joint space It is converted back to ^ref and distributed to each actuator 1. The acceleration reference value of each actuator 1 is realized by a robust control system based on the disturbance observer.

FIG. 8 shows the experimental results in the extraction mode. A human uses three fingers, the thumb, forefinger, and middle finger, to apply force to the three robot systems A to C to perform a transport operation while gripping an object (object). The time-position characteristics of the robot systems A to C at this time are shown in FIGS. In addition, the characteristics of the time-position response in the mode space transformed using the mode queuing matrix Q _n are shown in FIGS. Mode 1 means the center of gravity (COG), mode 2 means yawing, and mode 3 means internal operations. From the experimental results in the mode space, the position response of mode 1 is largely displaced first, and then the position response of mode 2 is largely displaced. Therefore, humans first perform the COG operation, and then perform the yawing operation. It can be confirmed. Also, since the hybrid ratio HR is set to 1 in all modes, human force feedback control is realized, and motion in the mode space can be extracted.

In this extraction mode, the acceleration disturbance m ^{·· ext} and the acceleration response value m ^{·· res} , which are the acceleration information of the mode space obtained by the mode acceleration conversion means 27 and 28, can be stored in the storage means 30. it can. Thereby, the motion extracted in the mode space can be stored and saved in the acceleration dimension. The storage means 30 stores the individual modes obtained by the interaction mode control unit 31 based on the acceleration disturbance m ^{·· ext} , the acceleration response value m ^{·· res,} and the hybrid ratio HR determined by the hybrid determination means 32. You may make it preserve | save the position response and force response in space as acceleration information converted into non-interference mode.

  On the other hand, in the reproduction mode, each robot system A to C automatically grasps an object, and the motion extracted in the extraction mode is read from the storage means 30 and can be reproduced by each robot system A to C. . Here, when a task command indicating that the reproduction mode has been selected is output to the hybrid ratio determining means 32, the hybrid ratio determining means 32 sets the hybrid ratio HR for mode 1 and mode 2 to 0 in the three non-interference modes described above. And the hybrid ratio HR for mode 3 is set to 1. This is because it is necessary to reproduce the position dimension for the gravity center mode of mode 1 and the yawing mode of mode 2 and to reproduce the force dimension for the internal mode of mode 3. The task code at this time is as follows.

Further, in order to realize the task code represented by Equation 13, the transmission elements C _i (s) (i = 1 to 3) of the acceleration control system in the mode space are set as follows.

In _Equation 14, K _pi (i = 1, 2) is a position control gain, and K _vi (i = 1, 2) is a speed control gain. K _f3 is a force servo gain in the internal force control mode.

At the same time, since the hybrid ratio HR of the modes 1 and 2 is both 0 and the hybrid ratio HR of the mode 3 is 1, the hybrid ratio determining means 32 is the mode 1 converted by the second mode acceleration converting means 28. , 2 acceleration response values m ₁ ^..res , m ₂ ^..res and the mode 3 acceleration response values m ₃ ^..ext converted by the first mode acceleration conversion means 27 are obtained by the interaction mode control unit 31. The task code selection means 51 is controlled so as to be input to the subtracters 52A to 52C. In the reproduction mode, the acceleration target values m ₁ ^{·· cmd} , m ₂ ^{·· cmd} , m ₃ ^{·· cmd} for each mode space obtained from the acceleration information stored and stored in the storage means 30 are subtracted by the interaction mode control unit 31. The interaction mode control unit 31 inputs the calculation results from the subtractors 52A to 52C to the transfer element C _i (s) (i = 1 to 3) described in the above equation 14, respectively. The acceleration reference values m ₁ ^{·· ref} , m ₂ ^{·· ref} , m ₃ ^{·· ref} in the mode space are calculated by multiplying them respectively. The acceleration reference value ^{_{m 1 ·· ref, m 2 ··}} ref, m 3 ·· ref is the reverse mode qualifiers over matrix _Q ^{3 -1} in the mode acceleration inverse transformation means 33, an acceleration reference value ^{x · ·} in the joint space It is converted back to ^ref and distributed to each actuator 1. The acceleration reference value of each actuator 1 is realized by a robust control system based on the disturbance observer.

  In order to confirm the effectiveness of the method proposed in this embodiment, a reproduction experiment using the conventional position control and a reproduction experiment using the interaction mode control were compared. The reproduction result by the conventional position control is shown in FIG. FIGS. 9A to 9C show the time-position characteristics in each of the robot systems A to C, and FIGS. 9D to 9F are diagrams in the mode space obtained by the mode-quarty matrix. The time-position response characteristics are shown. Further, FIG. 9G shows the time-force response in the internal force mode. From these results, it can be seen that the position error is almost zero due to the position servo in each of the robot systems A to C. However, since the value of the internal force is indefinite, the reproduction of the gripping motion is not guaranteed (see FIG. 9 (g)). Therefore, human motion cannot be reproduced only by conventional position control.

  On the other hand, the reproduction result by the proposed interaction mode control is shown in FIG. FIGS. 10A to 10C show the time-position characteristics in each of the robot systems A to C, and FIGS. 10D to 10F are views in the mode space obtained by the mode-quarty matrix. The time-position response characteristics are shown. From these results, it can be seen that the COG operation and the yawing operation are reproduced by the position servo in each mode. FIG. 10G shows the time-force response characteristics in the internal force mode. Here, the grip force command in the internal force mode is 1N. FIG. 10 (g) shows that the internal force can be controlled to be constant regardless of the COG operation or yawing operation. Therefore, it can be confirmed that the interaction mode control in this embodiment is useful in the extraction and reproduction of human skilled motion.

  As described above, the present embodiment proposes a method for extracting or reproducing a human skilld motion based on the interaction mode control. The interaction mode control here can extract the non-interacting mode of the robot system group using the mode queuing matrix, so it is very useful for extracting and reproducing the motion corresponding to the human five fingers. . In addition, a new control index called the hybrid ratio was defined as the ratio of the influence on the acceleration response value from the acceleration disturbance from other systems. As a result, a task code for describing a task in human motion can be established, and force control and position control can be handled uniformly on an acceleration basis.

  Next, a second embodiment of the present invention will be described with reference to FIGS. 11 to 17. Common portions to the first embodiment are denoted by the same reference numerals, and the description of the common portions is avoided to avoid duplication. Is omitted as much as possible.

  FIG. 11 shows a block diagram of the motion extraction mode in the motion acquisition system in the present embodiment. The second embodiment uses the master system 61 composed of a plurality of robot systems A to C and the slave system 62 composed of a plurality of robot systems A to C as the robot system group, and the skill of the skilled worker is obtained. We propose a control method that can be extracted or reproduced artificially. Each actuator 1 of the master system 61 and the slave system 62 arranged in the joint space is controlled by bilateral control based on acceleration control like the disturbance observer 11 described above. Bilateral control refers to the position control from the master system 61 to the slave system 62 and the control from the slave system 62 by performing control so that the position and force state are matched between the master system 61 and the slave system 62. This is a control method in which force control to the master system 61 is performed at the same time, and the control means for embodying it is well known and will not be described in detail here. By using such bilateral control, the action force and the reaction force of the operator can be separated from the force information that is bidirectional information governed by the “law of action / reaction”. Also, when storing and reproducing skills, as in the first embodiment, both position information and force information can be integrated by configuring a control system with acceleration dimensions for each non-interference mode based on interaction mode control. It can be expected to be applied to the development of tactile-based skill databases and skill simulations.

  The motion learning system shown in FIG. 11 includes three master systems 61 and slave systems 62 corresponding to three human fingers. Also in this experimental apparatus, a rod-type linear actuator is used as the actuator 1. Therefore, the work space and the joint space are equal, and the acceleration conversion means 25 and 26 and the acceleration reverse conversion means 34 as shown in FIG. 4 are not necessary.

  FIG. 12 is an example of an experimental apparatus for realizing the apparatus configuration shown in FIG. Each configuration is the same as that of the first embodiment, so that the description thereof will be omitted. However, in the extraction mode described later, a master system 61 operated by the operator 64 and a slave system that holds the object 47 in accordance with the operator. 62 are juxtaposed.

Returning to FIG. 11 again, the first mode acceleration conversion means 27A converts the acceleration disturbance x _Nm ^{·· ext} (N = A, B, C) of each robot system A to C constituting the master system 61 in the mode space. converts the acceleration disturbance ^{_{m im ·· ext (i = 1~3}} ), another first mode acceleration conversion means 27B is in the robot system A~C constituting the slave system 62 acceleration disturbance x _Ns ^{· · ext} (N = A, B, C) _is converted into acceleration disturbance m _is ^{·· ext} (i = 1 to 3) in the mode space. Similarly, the second mode acceleration conversion means 28A uses the acceleration response values x _Nm ^... (N = A, B, C) of the robot systems A to C constituting the master system 61 as the acceleration response values in the mode space. m converts to _im ^·· (i = _1~3), another second mode acceleration conversion means 28B is in the robot system A~C constituting the slave system 62 acceleration disturbance x _Ns ^·· (N = A, B, C) are converted into acceleration response values m _is ^... (I = 1 to 3) in the mode space.

Here, the acceleration response values of the master system 61 and the slave system 62 in the mode space are defined as m _m ^·· and m _S ^·· , respectively, and the equivalent acceleration of the external force of the master system 61 and the slave system 62 in the mode space is defined as Define m _m ^{·· ext} and m _S ^{·· ext} respectively. The goal of bilateral control can be described as follows.

The above formula 15 indicates that the sum of the equivalent accelerations of the external forces received by the master system 61 and the slave system 62 should be zero in order to realize the “law of action / reaction”. On the other hand, Equation 16 indicates that the deviation of the acceleration response should be zero in order to ensure synchronization between the master system 61 and the slave system 62. As described above, since the target of the bilateral control system is defined by the coordinate system of the sum and difference of the master system 61 and the slave system 62, the following equations 17 to 19 are used using the second-order mode queuing matrix Q2. It is possible to express them together as follows.

Here, m _c ^{· ·} and m _d ^{· ·} may represent the mode of the sum and difference of the acceleration response in the respective mode space, m _c ^{· · ext} and m _d ^{· · ext} is equivalent acceleration of external forces in the respective mode space Represents the sum and difference modes. That is, in the system of FIG. 11, by using the mode qualifier chromatography matrix Q ₂ of the number 19, the mode equivalent acceleration (acceleration disturbance) in the space m _im ^{· · ext,} the _m is ^{·· ext, ·} the sum m _ics ^{· ext} the difference m _id ^{· · ext} first acceleration calculating means 71A for calculating respectively the values of, 71B, and 71C, acceleration response m _im ^{· ·} in the modal ^space, the m _iS ^{· ·,} the sum m _ics a second acceleration calculating means 72 for calculating the values of ^... the difference m _id ^..., it is provided, respectively.

Further, in order to realize the formula 15, the following configuration is set by the following force servo with respect to the equivalent acceleration m _c ^{·· ext} of the external force in the sum mode.

Further, in the sum mode, the reaction force needs to be controlled according to human input. Therefore, in the extraction mode, the hybrid ratio HR is set to 1 in all spatial modes. That is, the hybrid ratio HR _c in the sum mode is expressed as follows.

Based on the above hybrid ratio, each transmission element C _ic (s) (i = 1 to 3) of the acceleration control system in the mode space is set as follows. The following K _fic (i = 1 to 3) represents a force control gain.

On the other hand, in the difference mode, constant value control is performed so that the deviation of the acceleration response is zero in order to realize Equation (16).

Here, C _p represents a position regulator. In the difference mode, it is necessary to compensate for disturbances and ensure position synchronization, so the hybrid ratio must be zero in all spatial modes. Accordingly, the hybrid ratio HR _d in the difference mode is expressed as follows.

Based on the hybrid ratio HR _d , the acceleration control system C _id (s) (i = 1 to 3) in the mode space is set as follows.

Here, K _fid (i = 1 to 3) represents a position control gain, and K _vid (i = 1 to 3) represents a speed control gain. By doing so, it is possible to design the force servo and the position regulator independently in the Cartesian coordinate system of sum and difference.

To calculate the above sum in the mode spatial modes of the acceleration reference value ^{_{m ic ·· (i = 1~3)}} , an acceleration reference value mode of a difference m _id ^{· ·} a (i = 1~3), this motion In the extraction mode, a sum mode control unit 74 and a difference mode control unit 75 are provided as interaction mode control units in the mode space. OR mode control unit 74 is calculated by the first acceleration calculating means 71, a subtractor equivalent acceleration m _ics ^{· · ext} of the external force in the mode of the sum of each mode space that is outputted through 77A～77C, the number 21 and respectively multiplied the described transfer element C _ic (s) and the the number 23, first integration means for calculating an acceleration reference value m _ics ^{· · ref} of the sum mode in the mode space shown in several 20 78A～ 78C is provided. The difference mode control unit 75, transfer element C of the acceleration response value m _id ^{· ·} in the mode of the difference between each mode space calculated by the second acceleration calculating means 72, described in the number 25 and number 27 There are provided second integrating means 79A to 79C for multiplying _id (s), respectively, to calculate the acceleration reference value m _id ^{·· ref} of the difference mode in the mode space shown in Formula 24.

The mode space, integrating force servo with position regulator acceleration dimension as the following equation, the acceleration reference value m _im ^{· · ref} modes space corresponding to the master system 61 and the slave system 62, m _IS ^· An acceleration reverse calculation means 81 for generating ^ref is provided.

Each equation of the above equation 28 can be combined as follows using the inverse matrix Q ₂ ⁻¹ of the second-order mode query matrix Q ₂ expressed by the equation 19.

Acceleration reference value m _im ^{· · ref} modes space corresponding to the master system 61 and the slave system 62, the m _IS ^{· · ref} is reverse mode in the mode acceleration inverse transform unit 33A, 33B qualifiers over matrix Q ₃ ^-1, The acceleration reference values x _Nm ^{·· ref} and x _Ns ^{·· ref} (N = A, B, C) in the joint space are respectively inversely converted and distributed to the actuators 1. The acceleration reference value of each actuator 1 is realized by a robust control system based on the disturbance observer, as in the first embodiment. The operation of the mode acceleration reverse conversion means 33A and 33B is expressed by the following formula.

FIG. 13 shows an experimental result in the motion extraction mode of the system having the above configuration. When an operator operates the master system 61 to apply an external force, the slave system 62 performs a task of gripping and moving the object 47 according to the operation. Master system 61 and the slave system 62 to time the actuator 1 is provided - shows the characteristic of position response in FIG. 13 (a) ~ (c) , the position response of each mode space using the mode qualifier chromatography matrix Q ₃ The respective conversions are shown in FIGS. 13 (d) to (f). In addition, in the mode space, the time-force response characteristic of the internal mode corresponding to the gripping force to the gripping object is shown in FIG.

  From this experimental result, it can be confirmed that the operator moves the object 47 while gripping the object 47 in the order of the center of gravity, yawing, and center of gravity. Further, by configuring the master system 61 and the slave system 62, it is possible to extract the gripping force of the gripped object operated by a human (operator). Furthermore, the position response between the master system 61 and the slave system 62 follows very well, and the “law of action / reaction” is artificially reproduced. Therefore, even in the system proposed in this embodiment, it is possible to extract the motion of three human fingers for each independent mode and analyze the skill.

Next, the motion reproduction mode of the second embodiment will be described. FIG. 14 shows a system configuration in which the motion in the mode space extracted in FIG. 11 is automatically reproduced by the slave system 62 installed in the joint space. Instead of the slave system 62, the master system 61 having the same configuration may be used. In FIG. 14, this is described as robot systems A to C. As described in the first embodiment, when realizing the object gripping task, the center of gravity mode and the yawing mode shown in FIG. 6 are reproduced in the position dimension, and the internal mode is reproduced in the force dimension. Therefore, the hybrid ratio HR _r in the reproduction mode can be expressed as follows.

Based on the hybrid ratio HR _r described above, each transfer element C _i (s) (i = 1 to 3) of the acceleration control system in the mode space is set as follows. This is the same as that shown in Equation 14 above.

Hybrid ratio determining means 32, the hybrid ratio HR _r is 0 mode 1, since the hybrid ratio HR _r of mode 3 is 1, converted mode in the second mode the acceleration conversion means 28 1,2 acceleration response values m ₁ ^{· ·,} and m ₂ ^{· ·,} acceleration response value m ₃ ^{· · ext} of the transformed mode 3 in the first mode the acceleration conversion means 27, a subtracter 77A~ interaction mode control unit 31 The task code selection means (not shown) is controlled individually so as to be input to 77C. In the reproduction mode, the acceleration target values m ₁ ^{·· cmd} , m ₂ ^{·· cmd} , m ₃ ^{·· cmd} for each mode space obtained from the acceleration information stored and stored in the storage means 30 are subtracted by the interaction mode control unit 31. The interaction mode control unit 31 inputs the calculation results from the subtractors 77A to 77C to the transfer element C _i (s) (i = 1 to 3) described by the equation 32. The acceleration reference values m ₁ ^{·· ref} , m ₂ ^{·· ref} , m ₃ ^{·· ref} in the mode space are calculated by multiplying them respectively. The acceleration reference value ^{_{m 1 ·· ref, m 2 ··}} ref, m 3 ·· ref is the reverse mode qualifiers over matrix _Q ^{3 -1} in the mode acceleration inverse transformation means 33, an acceleration reference value ^{x · ·} in the joint space It is converted back to ^ref and distributed to each actuator 1. The acceleration reference value of each actuator 1 is realized by a robust control system based on the disturbance observer.

  It should be noted that the interaction mode control unit 31 here can have the same configuration as the sum mode control unit 74 shown in FIG. However, the transfer elements set by the integrating means 78A to 78C are different between the motion extraction mode and the reproduction mode.

  Also in the present example, in order to confirm the effectiveness, a comparison between a reproduction experiment using the conventional position control and a reproduction experiment using the interaction mode control employed in this example was performed. Here too, the operator operates the master system 61, and the slave system 62 performs a task of gripping and moving the object 47 according to the operation. The reproduction result by the conventional position control is shown in FIG. FIGS. 15A to 15C show the time-position characteristics in each of the robot systems A to C, and FIGS. 15D to 15F show in the mode space obtained by the mode-quarty matrix. The time-position response characteristics are shown. On the other hand, FIG. 16 shows a reproduction result by the interaction mode control employed in this embodiment. FIGS. 16A to 16C show the time-position characteristics in each of the robot systems A to C, and FIGS. 16D to 16F are diagrams in the mode space obtained by the mode-quarty matrix. The time-position response characteristics are shown. When both are compared, the reproducibility of the proposed method of the present embodiment is also good with respect to the position of the center of gravity and the yawing position. Further, FIG. 17 shows the time-force response characteristics of the master system 61 and the slave system 62 extracted by the configuration shown in FIG. The characteristics of the time-force response by the position servo) and the characteristics of the time-force response in the interaction mode control proposed in this embodiment are shown in (a) to (c), respectively. As is apparent from this experimental result, the conventional method is based on position control, so the gripping force is indefinite, and the object 47 cannot be gripped with this. On the other hand, in the proposed method of the present embodiment, a constant gripping force is reproduced regardless of changes in the center of gravity position and yawing position, and the object 47 can be gripped satisfactorily.

  From these experimental results, it can be confirmed that in the proposed interaction mode control, the object gripping task is stored in, for example, three independent modes and reproduced.

  This embodiment proposes a motion learning system that extracts a human motion by using a master system 61 and a slave system 62 and artificially reproduces them. The motion learning system here is composed of two modes: a motion extraction mode configured by a bilateral control system and a motion reproduction mode configured by interaction mode control. In the motion extraction mode, it is possible to separate the action force and the reaction force of the operator from force information that is bidirectional information governed by the “law of action / reaction”. Also, to extract the non-interference mode defined by the Cartesian coordinate system corresponding to the human finger using the mode queuing matrix and analyze the skill for each human movement such as grasping, moving, and rotating the object. A motion extraction method was proposed. In the motion reproduction mode, human motion obtained in the extraction mode can be faithfully reproduced based on the interaction mode control. If the proposed method is applied to an object gripping task using, for example, three linear actuators, the motion consisting of three human fingers is extracted and stored, and the center of gravity and yawing movement and gripping force are automatically reproduced. It becomes possible.

  As described above, in each of the above-described embodiments, the first step of performing the acceleration control for making the control rigidity zero while maintaining the robustness is performed on the actuator 1 of each robot system arranged corresponding to the degree of freedom of movement in the real space; First, the equivalent acceleration of the external force applied to the actuator 1 and the acceleration response of the actuator 1 are converted into equivalent non-interference mode equivalent accelerations and acceleration responses, which are defined in an orthogonal coordinate system, using a mode quality matrix. And a third step of extracting a motion for the object 47 corresponding to each non-interference mode based on the equivalent acceleration and acceleration response in the non-interference mode obtained in the second step. A control method for motion learning system is proposed.

  As described above, since the acceleration control for reducing the control rigidity to zero without losing the robustness is performed on each actuator 1 of the plurality of robot systems, an operation involving contact with the environment can be realized. In addition, it is possible to newly define a non-interfering mode of the Cartesian coordinate system based on human movements and tasks using the mode queuing matrix, and each can extract independent motions. Extraction of motion corresponding to is easy.

  This includes a disturbance observer 11 as an acceleration control means functionally including the first step, and mode acceleration conversion means 27, 27A, 27B, 28, 28A, functionally including the second step. 28B and a motion learning system functionally including the third step (interaction mode control unit 31, hybrid ratio determining unit 32, sum mode control unit 74, difference mode control unit 75) Can also be realized.

  When extracting the motion, the robot system includes a master system 61 and a slave system 62. In the first step, each actuator 1 of the master system 61 and the slave system 62 is controlled by the acceleration control. Preferably, the acceleration control means is configured to control by bilateral control based on the above, or to control each actuator 1 of the master system 61 and the slave system 62 by bilateral control based on the acceleration control.

  In this way, the action force and reaction force of the operator can be separated from the haptic information, which is bidirectional information governed by the “law of action / reaction”. The reaction force can be extracted.

  Further, it is preferable to further include a fourth step of storing the acceleration information converted into the non-interference mode, or to further include storage means 30 for storing the acceleration information converted into the non-interference mode.

  If it carries out like this, it will become possible to preserve | save the extracted motion and to read the acceleration information converted into non-interference mode in another place, for example.

  Further, for each non-interference mode, a ratio of the acceleration response to the equivalent acceleration of the external force in the non-interference mode (hybrid ratio HR) is determined, and an acceleration target value based on the acceleration information stored in the fourth step; The acceleration reference value is calculated from the acceleration control system in the non-interference mode set based on the hybrid ratio HR using the deviation from the equivalent acceleration or acceleration response of the external force in the non-interference mode obtained in the second step as an input. And a sixth step of converting an acceleration reference value in the non-interference mode into an acceleration reference value for each robot system using an inverse matrix of the mode-quarty matrix.

  In this way, it can be reproduced by the same or different robot system based on the acceleration information that has been extracted and stored in the non-interference mode, and the reproducibility of motion is ensured. In particular, since the control system is configured with an acceleration dimension for each non-interference mode, position information and force information can be integrated. If the hybrid ratio is set to 1 or 0, each non-interference mode It is possible to reproduce in the position dimension or force dimension, and reflect it in the motion of the robot system in real space.

  For each non-interference mode, this determines the ratio of the acceleration response to the equivalent acceleration of the external force in the non-interference mode (hybrid ratio HR), and the acceleration target value based on the acceleration information stored in the storage means 30 In the non-interference mode set based on the hybrid ratio HR ratio with the deviation from the equivalent acceleration or acceleration response of the external force in the non-interference mode obtained by the mode acceleration conversion means 27, 27A, 27B, 28, 28A, 28B as an input. Mode acceleration calculation means (interaction mode control unit 31, hybrid ratio determination unit 32, sum mode control unit 74, difference mode control unit 75) for calculating the acceleration reference value from the acceleration control system, and an inverse matrix of the mode quary matrix Is used to convert the acceleration reference value in the non-interference mode to the acceleration reference value for each robot system. It can also be realized by a configuration further including replacement means 33, 33A, 33B.

  In addition, the effect of friction can be mechanically removed by using a linear actuator that performs linear driving as the actuator.

  In addition, this invention is not limited to the said Example, A various deformation | transformation implementation is possible in the range of the summary of this invention.

  The motion learning system and the control method proposed in this embodiment are highly novel because they extract the skill of an expert using an interaction mode control or a master / slave system and artificially reproduce it. In particular, since the master / slave system is controlled by a bilateral control system based on acceleration control, it is possible to realize a stable contact operation with an unknown actual environment and an integrated motion with multiple degrees of freedom. Furthermore, with the introduction of a mode-quarty matrix, for example, a non-interference mode defined by the Cartesian coordinate system corresponding to the human five fingers is extracted, and skills are saved and stored for each human action such as grasping, moving, and rotating objects. It is possible to reproduce. When storing and reproducing skills, it is possible to integrate both position information and force information by configuring a control system with acceleration dimensions for each non-interference mode. I can expect.

It is a block diagram showing an example of the robot system applicable to the motion acquisition system in 1st Example of this invention. FIG. 2 is a block diagram of an acceleration control system using a disturbance observer equivalent to FIG. 2 is a control block diagram of a disturbance observer incorporated in FIG. It is a block diagram which shows the general structure which materialized interaction mode control same as the above. It is a perspective view which shows an example of the experimental apparatus which implement | achieves the apparatus structure shown in FIG. 4 same as the above. It is explanatory drawing which shows the object holding | grip task with three fingers in mode space same as the above. FIG. 2 is a block diagram of a motion learning system that realizes extraction and reproduction control of a human object gripping motion based on interaction mode control. It is a graph of each characteristic which shows the extraction result of the motion by interaction mode control. It is a graph of each characteristic which shows the reproduction result of the motion by the conventional position control. It is a graph of each characteristic which shows the reproduction result of the motion by interaction mode control. It is a block diagram of the motion extraction mode in the motion acquisition system which shows 2nd Example of this invention. It is a perspective view which shows an example of the experimental apparatus which implement | achieves the apparatus structure shown in FIG. 11 same as the above. It is a graph of each characteristic which shows the extraction result of a motion same as the above. It is a block diagram of a motion acquisition system suitable for automatically reproducing the motion extracted in the motion extraction mode. It is a graph of each characteristic which shows the reproduction result regarding the position response of the motion by the conventional position control. It is a graph of each characteristic which shows the reproduction result regarding the position response of the motion based on the experimental result of a present Example same as the above. It is a graph which shows the reproduction result regarding the force response of an internal mode.

Explanation of symbols

11 Disturbance observer (acceleration control means)
27, 27A, 27B, 28, 28A, 28B Mode acceleration conversion means
30 Storage means
31 Interaction mode controller (motion extraction means, mode acceleration calculation means)
32 Hybrid ratio determination means (motion extraction means, mode acceleration calculation means)
33, 33A, 33B mode acceleration reverse conversion means
74 Sum mode control unit (motion extraction means, mode acceleration calculation means)
75 Difference mode controller (motion extraction means, mode acceleration calculation means)

Claims (10)

A first step of performing acceleration control to zero control rigidity while maintaining robustness for each actuator of a plurality of robot systems arranged corresponding to the degree of freedom of movement in real space;
The equivalent acceleration of external force to each actuator and the acceleration response of each actuator are converted into the equivalent acceleration and acceleration response of external force in independent non-interfering mode defined by orthogonal coordinate system using mode queuing matrix A second step of:
A third step of extracting a motion for the object corresponding to each of the non-interference modes based on the equivalent acceleration and acceleration response of the external force in the non-interference mode obtained in the second step. A control method of the featured motion acquisition system.   The robot system includes a master system and a slave system, and the actuators of the master system and the slave system are controlled by bilateral control based on the acceleration control in the first step. Item 8. A motion learning system control method according to Item 1.   The method of controlling a motion learning system according to claim 1, further comprising a fourth step of storing the acceleration information converted into the non-interference mode. For each non-interference mode, the ratio of the acceleration response to the equivalent acceleration of the external force in the non-interference mode is determined, and the acceleration target value based on the acceleration information stored in the fourth step is obtained in the second step. A fifth step of calculating an acceleration reference value from the acceleration control system in the non-interference mode set based on the ratio, with the deviation from the equivalent acceleration or acceleration response of the external force in the non-interference mode as an input;
The method further comprises a sixth step of converting an acceleration reference value in the non-interference mode into an acceleration reference value for each of the robot systems using an inverse matrix of the mode quality matrix. The motion learning system control method described.   The method of controlling a motion learning system according to claim 1, wherein the actuator is a linear actuator. Acceleration control means for performing acceleration control to zero control rigidity while maintaining robustness for each actuator of a plurality of robot systems arranged corresponding to the degree of freedom of movement in real space;
A mode in which the equivalent acceleration of external force to each actuator and the acceleration response of each actuator are converted into equivalent non-interference mode equivalent accelerations and acceleration responses, each defined by a rectangular coordinate system using a mode-quarty matrix. Acceleration conversion means;
And a motion extraction means for extracting a motion for an object corresponding to each of the non-interference modes based on an equivalent acceleration and an acceleration response in the non-interference mode obtained by the mode acceleration conversion means. To learn motion.   The robot system is composed of a master system and a slave system, and the acceleration control means controls each actuator of the master system and the slave system by bilateral control based on the acceleration control. The motion learning system according to claim 6.   8. The motion learning system according to claim 6, further comprising storage means for storing acceleration information converted into the non-interference mode. For each non-interference mode, the ratio of the acceleration response to the equivalent acceleration of the external force in the non-interference mode is determined, and the acceleration target value based on the acceleration information stored in the storage means and the mode acceleration conversion means are obtained. A mode acceleration calculation means for calculating an acceleration reference value from an acceleration control system in the non-interference mode set based on the ratio, with the deviation from the equivalent acceleration or acceleration response of the external force in the non-interference mode as an input,
The mode acceleration reverse conversion means for converting an acceleration reference value in the non-interference mode into an acceleration reference value for each robot system by using an inverse matrix of the mode quary matrix. 8. Motion acquisition system according to 8.   The motion learning system according to claim 6, wherein the actuator is a linear actuator.