JP3555945B2 - Motion editing device and motion editing method for robot device, robot device, control method for robot device, and computer program - Google Patents

Description

  The present invention relates to a motion editing device and a motion editing method for supporting creation and editing of a motion describing a predetermined motion pattern of a robot, and a computer program, and more particularly to a robot device that performs various operations with movable legs. And a motion editing method, a robot device, a control method of a robot device, and a computer program for the same.

  More specifically, the present invention relates to a motion editing device and a motion editing method for a robot device that supports editing of an operation pattern while considering the feasibility on a real machine, a robot device, a control method of the robot device, and a computer The present invention relates to a program, and more particularly, to a motion editing device and a motion editing method for a robot device that corrects an edited motion while confirming an operation on a real machine, a robot device, a control method of the robot device, and a computer program.

  A mechanical device that performs a motion resembling a human motion using an electric or magnetic action is called a “robot”. It is said that the robot is derived from the Slavic word "ROBOTA (slave machine)". In Japan, robots began to spread from the late 1960's, but most of them were industrial robots (industrial robots) such as manipulators and transfer robots for the purpose of automation and unmanned production work in factories. Met.

  Recently, research and development on legged mobile robots that imitate the body mechanisms and movements of animals such as humans and monkeys that walk upright on two legs have been progressing, and expectations for their practical use have increased. Leg-based movement with two feet standing upright is more unstable than the crawler type, four-legged or six-legged type, and makes posture control and walking control difficult, but walking with irregularities on the work route such as uneven terrain and obstacles It is excellent in that it can realize flexible moving work, for example, it can respond to discontinuous walking surfaces such as surfaces, stairs and ladders.

  In addition, a legged mobile robot that reproduces a human biological mechanism or motion is particularly called a “humanoid” or “humanoid” robot. The humanoid robot can perform, for example, life support, that is, support for human activities in various situations in a living environment and other daily lives.

  Most of the working space and living space of human beings are formed according to the human body mechanism and behavior style of bipedal upright walking, and the current mechanical system using wheels and other driving devices as moving means will move. There are many barriers. Therefore, in order for the mechanical system, that is, the robot, to perform various human tasks and penetrate deep into the human living space, it is preferable that the movable range of the robot is almost the same as that of a human. This is the reason why practical use of the legged mobile robot is greatly expected.

  Recent legged mobile robots have high information processing ability, and can be regarded as a kind of computer system. In other words, a motion pattern realized on the robot, or a highly complex series of motion sequences or motions constituted by a combination of a plurality of basic motion patterns is constructed by a task similar to computer programming. .

  It is indispensable that a large number of motion data for operating an actual machine be widely used in order for the robot body to be widely used. Therefore, construction of a development environment for performing motion editing for a robot is strongly desired.

  In the future, it is expected that robots will penetrate not only into the industrial world but also into ordinary households and everyday life. In particular, for products that pursue entertainment, it is strongly desired that choreographers and designers can create motion content even if they have advanced knowledge of motion control in addition to computers and computer programming. It is expected that there are many cases where general consumers purchase and use robots. Even for such a general user, it is preferable to provide a tool for supporting creation and editing of the operation sequence of the robot relatively easily and efficiently by interactive processing, that is, a motion editing system.

  The robot is composed of a plurality of control points such as joints. Therefore, the robot operation can be edited by inputting the position and speed (joint angle and its angular speed) at each control point. In this respect, it is analogous to character animation in computer graphics. However, there is naturally a difference between the operation in the virtual space and the operation of the actual device. In the case of a legged mobile robot, a desired operation cannot be performed simply by driving the joint angle, and the legged work must be continued without falling. In other words, it is a prerequisite for realizing the desired operation that the operation on the actual device is confirmed and that the aircraft maintains the posture stability during the execution of the motion.

  In many cases, the posture stability control of a legged mobile robot uses a ZMP stability discrimination criterion of searching for a point where the moment becomes zero inside a supporting polygon formed by a sole and a ground contact point (for example, See Non-Patent Document 1). In the case of a two-legged mobile robot, since the supporting polygon is extremely small, the posture stabilization control is particularly difficult.

There is already a motion editing system that inputs the indicated values at each control point of the aircraft on the screen and forms the motion of the robot.However, however, the posture stability when operating the actual machine with the edited motion is checked, There is no system that modifies a desired motion so as to be stabilized. In the case of a pre-assembled motion, the attitude stability of the aircraft cannot be maintained, and if the motion itself cannot be executed, the motion is not substantially edited.
"Legged Locomotion Robots" by Miomir Vukobratovic (Ichiro Kato, "Walking Robots and Artificial Feet" (Nikkan Kogyo Shimbun))

  SUMMARY OF THE INVENTION An object of the present invention is to provide a motion editing apparatus and a motion editing method for an excellent robot apparatus, which can support the editing of a motion pattern while considering the feasibility on a real machine, a robot apparatus, and control of the robot apparatus. It is to provide a method, as well as a computer program.

  A further object of the present invention is to provide a motion editing apparatus and a motion editing method for an excellent robot apparatus, which can correct an edited motion while confirming an operation on a real machine, a robot apparatus and a control method of the robot apparatus, And to provide a computer program.

The present invention has been made in view of the above problems, and a first aspect of the present invention is a motion editing apparatus or a motion editing method for a robot apparatus having a plurality of joint degrees of freedom and sensors for measuring an external environment. And
A data input unit or step for inputting motion data;
A data reproducing unit or step for reproducing motion data on an actual device;
A sensor information acquisition unit or step for acquiring sensor information from the sensor that is reproducing motion data;
A motion evaluation unit or step for evaluating a motion based on the acquired sensor information,
A motion correcting unit or step for correcting motion data based on the evaluation result;
A motion editing device or a motion editing method for a robot device, comprising:

  According to the motion editing apparatus or the motion editing method for a robot device according to the first aspect of the present invention, by acquiring sensor information of a real robot through communication, a motion considering a response of the real robot is created. Can be. Further, when the motion data created in the reference environment is executed in a different environment such as a residential environment, it can be determined whether or not the actual robot is operating as intended.

  Here, in the motion editing device or the motion editing method for a robot device according to the first aspect of the present invention, the motion information satisfying an evaluation criterion in the motion evaluation unit or the step is acquired by the sensor information acquisition unit. A motion data output unit or step for outputting as motion data with reference data to which reference data such as sensor information is added may be further provided.

  The reference data referred to here includes sensor information and operation information to be obtained during the actual reproduction of the motion data. Specifically, the measured value at the time of executing the motion for the angle command value for each joint, the combination of the measured value of each sensor at the time of executing the motion and the measured value after filtering the sensor output, the left and right soles The reference data includes the ZMP trajectory corrected by the stabilization control at the time of executing the motion with respect to the target ZMP trajectory at the time of editing, the measured value at the time of executing the motion with respect to the target value at the time of editing the floor reaction force sensor, and the like.

  Even if the motion of the robot described by the motion data is the same, the output of each sensor differs depending on the external environment or the work environment. For example, even if the walking pattern is the same, the output value of the sensor is different on a slope, on a gravel on which the road surface moves, or on a carpet with long hairs. By embedding the reference data in the motion data, there is an advantage that motion data corresponding to a specific working environment (or a robot use form) can be described.

  The motion data output unit or step includes: joint angle information composed of a combination of an angle command value for each joint and a measurement value at the time of motion execution; a target value at the time of motion editing for each sensor; and a measurement at the time of motion execution. Posture information consisting of a combination of values and sensor values after filtering the sensor output, a combination of the target ZMP trajectory when editing the left and right soles and the ZMP trajectory corrected by the stabilization control when executing the motion Trajectory information and / or contact information composed of a combination of a ZMP trajectory information consisting of a target value at the time of editing the floor reaction force sensor and a measurement value at the time of executing a motion as motion data with reference data. You may make it output.

  Further, the motion evaluation unit or the step may evaluate the followability when executed on a real robot in a time series or a frequency domain. When the evaluation is performed in the frequency domain, there is an advantage that a frequency-specific band can be limitedly handled, for example, by removing a natural frequency of a mechanical system.

  Further, the motion evaluation unit or the step obtains the actuator torque value and the number of rotations when executed by the real robot in a time series, compares the obtained data with an NT curve indicating the characteristics of the actuator, and determines the limit torque of the actuator. Alternatively, it may be evaluated whether there is an operation exceeding the limit rotation speed.

  Further, the motion evaluation unit or the step may calculate a deviation between a posture sensor value or a ZMP trajectory planned at the time of motion editing and a sensor value or a ZMP trajectory when executed by a real robot, and evaluate the posture. Good.

  Further, the motion evaluation unit or the step may calculate a deviation between a posture at the time of editing the motion and a measured value when the motion is executed on the real robot, and may evaluate the contact with the ground and / or the contact. .

  Further, the motion evaluation unit or the step may calculate the degree of improvement of the measured value for the motion corrected by the previous evaluation and evaluate the degree of achievement of the correction.

  Further, the motion evaluation unit or the step may calculate the influence of the actuator torque, the ZMP deviation and the acceleration due to the impact, and evaluate the impact due to the external contact.

  Further, the motion correction unit or the step may correct an angle instruction value to the actuator and / or correct a control parameter of the actuator based on the actuator responsiveness evaluation result.

  Further, the motion correction unit or the step may correct the motion data by a posture stabilization process based on the evaluation result of the actuator torque.

  Further, the motion correcting section or the step may correct the motion data by a posture stabilizing process based on the evaluation result of the touchdown and / or the contact.

  Further, the motion correction unit or the step may correct the motion data by a posture stabilization process in consideration of the external contact, based on the evaluation result of the impact due to the external contact.

  Further, the data reproducing unit or step may extract only an arbitrary range of the motion data and reproduce the motion data on an actual device.

  When the motion data is reproduced on the actual device, only an arbitrary range of the motion data is extracted and reproduced on the actual device, so that the efficiency of the motion editing work can be improved.

  In this case, the data reproducing unit or the step sets a start time in the motion data, calculates a dynamic posture at the start time, generates a transient motion having the dynamic posture at the start time as an end point, A motion is reproduced on a real machine using the transient motion, a stop time in the motion data is set, a dynamic posture at the stop time is calculated, and a transient motion starting from the stop posture is generated. Then, the operation of the actual machine may be stopped using the transient motion.

  A motion is composed of a chronological combination of two or more poses. When a dynamic motion (continuous dynamic motion) using acceleration positively and continuously is executed, the motion is executed from the middle. However, it is impossible to make an evaluation. Therefore, in the present invention, such continuous dynamic motion can be reproduced and stopped from an arbitrary time, so that the time required for motion evaluation is extremely reduced.

According to a second aspect of the present invention, there is provided a robot apparatus having a plurality of degrees of freedom of joints,
A data input unit for inputting motion data describing the device operation including each of the joints,
An operation control unit for reproducing motion data on a real machine,
A sensor information acquisition unit that acquires sensor information from the sensor that is reproducing motion data,
The data input unit, when inputting the motion data, combines the sensor data to be obtained from the sensor information acquisition unit or the reference data including the operation information to be obtained from the operation control unit when executing the motion data. Get
The operation control unit, the sensor information or the operation information described in the reference data, the sensor information obtained from the sensor information acquisition unit during the actual reproduction of the motion data or the operation information obtained from the operation control unit, And determines whether or not the reproduction of the motion data can be continued based on the comparison result.
A robot apparatus characterized in that:

  Even if the motion of the robot described by the motion data is the same, the output of each sensor differs depending on the external environment or the work environment. For example, even if the walking pattern is the same, the output value of the sensor is different on a slope, on a gravel on which the road surface moves, or on a carpet with long hairs. By embedding the reference data in the motion data, there is an advantage that motion data corresponding to a specific working environment (or a robot use form) can be described.

According to the robot apparatus according to the second aspect of the present invention, the reference data included in the motion data includes various reference data in a plurality of external environments such as different road environments other than the standard environment. In this case, the current external environment can be estimated by comparing the deviation between each reference data and the actual measurement data at the time of reproducing the motion data. Further, after estimating the external environment, it is possible to determine whether or not the reproduction of the motion data can be continued as it is according to the external environment.
In the past, when the aircraft became unstable due to the execution of motion data, there was no alternative but to start adaptation processing for falls and other abnormal situations. Can be stopped, and a motion appropriate for the environment can be reproduced adaptively by estimating the environment.

According to a third aspect of the present invention, a motion editing process for a robot apparatus having a plurality of joint degrees of freedom and a sensor for measuring an external environment is described in a computer-readable format so as to be executed on a computer system. Computer program,
A data input step for inputting motion data;
A data playback step for playing back the motion data on a real machine;
A sensor information acquisition step of acquiring sensor information from the sensor that is reproducing motion data;
A motion evaluation step of evaluating a motion based on the acquired sensor information;
A motion correction step of correcting motion data based on the evaluation result;
A computer program characterized by comprising:

  A computer program according to a third aspect of the present invention defines a computer program described in a computer-readable format so as to realize a predetermined process on a computer system. In other words, by installing the computer program according to the third aspect of the present invention into a computer system, a cooperative action is exerted on the computer system, and the leg type according to the first aspect of the present invention is provided. The same operation and effect as those of the motion editing device or the motion editing method for the mobile robot can be obtained.

  ADVANTAGE OF THE INVENTION According to the present invention, a motion editing apparatus and a motion editing method for an excellent robot apparatus for a robot apparatus which supports editing of a motion pattern while considering the feasibility on a real machine, a robot apparatus and a robot apparatus A control method and a computer program can be provided.

  Further, according to the present invention, a motion editing apparatus and a motion editing method for an excellent robot apparatus, which can correct an edited motion while confirming an operation on a real machine, a robot apparatus and a control method of the robot apparatus , As well as computer programs.

  Further objects, features, and advantages of the present invention will become apparent from more detailed descriptions based on embodiments of the present invention described below and the accompanying drawings.

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

A. Mechanical Configuration of Legged Mobile Robot FIGS. 1 and 2 show the external configuration of a legged mobile robot 100 to be subjected to motion editing by the motion editing system according to the present invention. The legged mobile robot 100 is called a “humanoid” or a “humanoid”. As shown, the legged mobile robot 100 performs a torso, a head, left and right upper limbs, and a legged movement. The left and right legs are configured to perform the operation of the fuselage by a control unit (not shown) built in the fuselage, for example.

  Each of the left and right lower limbs includes a thigh, a knee joint, a shin, an ankle, and a foot, and is connected at a substantially lowermost end of the trunk by a hip joint. Each of the left and right upper limbs is composed of an upper arm, an elbow joint, and a forearm, and is connected at the left and right side edges above the trunk by a shoulder joint. The head is connected to the center of the uppermost end of the trunk by a neck joint.

  The control unit includes a controller (main control unit) that processes drive control of each joint actuator constituting the legged mobile robot 100 and processes an external input from each sensor (described later) and a power supply circuit and other peripheral devices. It is the case which did. The control unit may include a communication interface and a communication device for remote control.

  The legged mobile robot 100 configured as described above can realize bipedal walking by controlling the whole body cooperatively by the control unit. Such bipedal walking is generally performed by repeating a walking cycle divided into the following operation periods. That is,

(1) Right leg lifted, left leg supports single leg (2) Right leg touches both legs support period (3) Left leg lifts, right leg supports single leg period (4) Left leg touches left leg Support period

  The walking control in the legged mobile robot 100 is realized by planning the target trajectory of the lower limb in advance and correcting various planned trajectories in each of the above periods.

  In general, the attitude stabilization control of the fuselage, including the correction of the trajectory of the walking motion, is performed by interpolation calculation using a fifth-order polynomial so that the position, velocity, and acceleration for reducing the deviation from ZMP are continuous. . ZMP (Zero Moment Point) is used as a criterion for determining walking stability.

  The stability discrimination standard based on ZMP is based on the principle of “Dallambert” that gravity and inertia force from the walking system to the road surface and these moments balance the floor reaction force and the floor reaction force moment as a reaction from the road surface to the walking system. based on. As a consequence of the mechanical inference, there is a point where the pitch axis and roll axis moments become zero, that is, ZMP, inside the support polygon (that is, the ZMP stable area) formed by the sole and the road surface.

  FIG. 3 schematically shows a degree of freedom configuration of the legged mobile robot 100 used in the embodiment of the present invention.

  As shown in the figure, the robot has limbs attached to the body, and has a shoulder joint pitch axis, shoulder joint roll axis, upper arm yaw axis, elbow joint pitch axis, forearm yaw axis, wrist roll axis, and wrist pitch axis. It has left and right arms with degrees of freedom, and left and right legs with six degrees of freedom: hip joint yaw axis, hip joint roll axis, hip joint pitch axis, knee pitch axis, ankle pitch axis, and ankle roll axis.

  These joint degrees of freedom are actually realized by an actuator / motor. In the present embodiment, a small AC servo actuator of a type directly connected to a gear and of a type in which a servo control system is integrated into a single motor and mounted in a motor unit is mounted. This type of AC servo actuator is disclosed, for example, in Japanese Patent Application Laid-Open No. 2000-299970 (Japanese Patent Application No. 11-33386) which has already been assigned to the present applicant.

  The body is equipped with an acceleration sensor A1 and a gyro G1. At the four corners of the left and right soles, a uniaxial load cell (F1 to F8) for detecting a floor reaction force in a vertical direction of the sole and an infrared distance measuring sensor (D1 to D8) for measuring a distance to the floor are respectively provided. Four are attached. Further, an acceleration sensor (A2, A3) and a gyro (G2, G3) are attached to the center of the left and right soles, respectively.

  Although not shown in FIG. 3 in order to prevent the drawings from being complicated, the head unit is moved relative to the torso with respect to the neck joint yaw axis, the first and second neck joint pitch axes, and the neck joint roll axis. The degree of freedom of the joint around each axis is provided. In addition, the trunk has joint degrees of freedom around each of the trunk roll axis and the physique pitch axis.

B. Control System Configuration of Legged Mobile Robot FIG. 4 schematically shows a control system configuration of the legged mobile robot 100. As shown in the figure, the legged mobile robot 100 has mechanism units 30, 40, 50R / L and 60R / L representing human limbs and adaptive control for realizing a cooperative operation between the mechanism units. (Where R and L are suffixes indicating right and left, respectively. The same applies hereinafter).

  The operation of the entire legged mobile robot 100 is totally controlled by the control unit 80. The control unit 80 transmits and receives data and commands to and from a main control unit 81 composed of main circuit components (not shown) such as a CPU (Central Processing Unit) and a memory, and a power supply circuit and each component of the robot 100. A peripheral circuit 82 including an interface (both not shown) is provided.

  In realizing the present invention, the installation location of the control unit 80 is not particularly limited. In FIG. 4, it is mounted on the trunk unit 40, but may be mounted on the head unit 30. Alternatively, the control unit 80 may be provided outside the legged mobile robot 100 to communicate with the body of the legged mobile robot 100 by wire or wirelessly.

Each joint degree of freedom in the legged mobile robot 100 shown in FIG. 3 is realized by the corresponding actuator motor M. That is, the head unit 30 includes a neck joint yaw axis actuator M ₁ , a neck joint pitch axis actuator M _2A that expresses each of a neck joint yaw axis, first and second neck joint pitch axes, and a neck joint roll axis. M _2B and a neck joint roll axis actuator M ₃ are provided.

Further, the trunk unit 40 is provided with a trunk pitch axis actuator M ₁₁ and a trunk roll axis actuator M ₁₂ that represent each of a trunk pitch axis and a trunk roll axis.

The arm unit 50R / L is subdivided into an upper arm unit 51R / L, an elbow joint unit 52R / L, and a forearm unit 53R / L. The shoulder joint pitch axis, shoulder joint roll axis, and upper arm yaw axis , The shoulder pitch axis actuator M ₄ , the shoulder roll axis actuator M ₅ , the upper arm yaw axis actuator M ₆ , and the elbow joint pitch that represent each of the elbow joint pitch axis, the elbow joint yaw axis, the wrist roll axis, and the wrist joint pitch axis An axis actuator M ₇ , an elbow joint yaw axis actuator M ₈ , a wrist joint roll axis actuator M ₉ , and a wrist joint pitch axis actuator M ₁₀ are provided.

The leg unit 60R / L is subdivided into a thigh unit 61R / L, a knee unit 62R / L, and a shin unit 63R / L. The hip joint yaw axis, the hip joint pitch axis, and the hip joint roll axis , knee joint pitch axis, ankle joint pitch axis, a hip joint yaw axis actuator M ₁₃ representing the respective ankle joint roll axis, a hip joint pitch axis actuator M _14, hip joint roll axis actuator M _15, knee joint pitch axis actuator M _16, ankle A joint pitch axis actuator M ₁₇ and an ankle joint roll axis actuator M ₁₈ are provided.

The actuators M ₁ , M ₂ , M ₃ ... Used for the joints are more preferably small-sized AC servo actuators of the type directly connected to a gear and having a single-chip servo control system and mounted in a motor unit (described above). ).

  Sub-control units 35, 45, 55, and 65 for actuator drive control are provided for each mechanism unit such as the head unit 30, the trunk unit 40, the arm unit 50, and each leg unit 60.

  An acceleration sensor A1 and a posture sensor G1 including a gyro sensor and the like are provided on the trunk 40 of the body. The acceleration sensor A1 is arranged, for example, in each of the X, Y, and Z axis directions. By arranging the acceleration sensor A1 on the waist of the body, the waist, which is a part where the mass operation amount is large, is set as the control target point, and the posture and acceleration at that position are directly measured, and the posture stabilization based on ZMP is performed. Control can be performed.

  In addition, floor reaction force sensors F1 to F4, F5 to F8, acceleration sensors A2 and A3, and posture sensors G2 and G3 are disposed on the legs 60R and 60L, respectively. The floor reaction force sensors F1 to F8 are configured by, for example, mounting a pressure sensor on the sole, and can detect whether the sole has landed on the basis of the presence or absence of the floor reaction force. The acceleration sensors A2 and A3 are arranged at least in the X and Y axial directions. By arranging the acceleration sensors A2 and A3 on the left and right feet, the ZMP balance equation can be assembled directly with the feet closest to the ZMP position.

  When the acceleration sensor is placed only on the waist, which is a part where the mass operation amount is large, only the waist becomes the direct control target point, and the state of the foot is relatively calculated based on the calculation result of this control target point. It is necessary to satisfy the following conditions between the foot and the road surface.

(1) The road surface does not move under any force or torque.
(2) The coefficient of friction against translation on the road surface is sufficiently large and no slip occurs.

  On the other hand, in the present embodiment, a reaction force sensor system (such as a floor reaction force sensor) that directly applies ZMP and force is provided on a foot portion that is a contact portion with a road surface, and local coordinates used for control and the coordinates thereof are used. The acceleration sensor for directly measuring is provided. As a result, the ZMP balance equation can be directly assembled with the foot closest to the ZMP position, and more precise posture stabilization control that does not depend on the preconditions described above can be realized at high speed. As a result, even on a gravel or a carpet with long fluff that the road surface moves when force or torque is applied, or on a residential tile where slippage tends to occur due to a lack of sufficient friction coefficient for translation, Stable walking (exercise) can be guaranteed.

  The main control unit 80 can dynamically correct the control target in response to the outputs of the sensors A1 to A3, G1 to G3, and F1 to F8. More specifically, a whole body motion pattern in which the upper limb, the trunk, and the lower limb of the legged mobile robot 100 are driven in a coordinated manner by performing adaptive control on each of the sub-control units 35, 45, 55, and 65. To achieve.

The whole body motion of the robot 100 on the body includes setting of a foot motion, a ZMP trajectory, a trunk motion, an upper limb motion, a waist height, and the like. Transfer to the sections 35, 45, 55, 65. Then, in each of the sub-control section 35, 45 ... interprets the command received from the main control unit 81 outputs a drive control signal to each actuator _{M 1, M 2, M 3} .... Here, “ZMP” refers to a point on the floor where the moment due to the floor reaction force during walking becomes zero, and “ZMP trajectory” refers to, for example, ZMP during the walking operation of the robot 100. Means a moving trajectory (described above).

C. Motion Editing System FIG. 5 schematically shows a processing flow in the motion editing system according to one embodiment of the present invention.

  First, a user edits motion data for the legged mobile robot 100 offline on a motion editing system such as a personal computer on which a motion editing application is installed (step S1).

  Here, the motion data can be edited by, for example, combining two or more poses (postures) of the robot 100 in time series. The pose of the robot 100 is described by the displacement amount of each joint angle. The motion data is composed of the displacement amount of each joint angle, its speed, acceleration, and the like. For example, motion data can be constituted by the following element data.

t1, t2, t3, t4, t5: time p: type of joint or sensor d: joint angle x, y, z, Yaw, Roll, Pitch: position and posture at joint g: target servo gain tr: type of interpolation fx , Fy, fz: forces ax, ay, az: accelerations vx, vy, vz: velocity

  The motion data itself is suitably edited, for example, by using the motion editing apparatus described in Japanese Patent Application No. 2002-73423 (JP-A-2003-266347) which has already been assigned to the present applicant. be able to.

  Next, the created motion data is installed in the legged mobile robot 100, and the operation is confirmed using the actual machine (step S2). Instead of installing the motion data on the actual device, the data may be distributed to the actual device in a streaming format.

  At this time, the motion data itself is time-series data including continuation of motion, that is, displacement and speed of each joint angle, and is sometimes long. In the present embodiment, the efficiency of the motion editing work is improved by extracting only an arbitrary range of the motion data and reproducing it on an actual device. This will be described in detail later.

  Next, when an arbitrary range of the motion data is reproduced using the actual machine, the output from each sensor installed on the machine, that is, the sensor information is transferred to the motion editing system (step S3). The sensor information referred to here includes, in addition to the rotation signal from the encoder provided in each joint actuator, the acceleration sensor A1 and the gyro sensor G1, which are provided in the trunk unit 40 (substantially the center of gravity of the body). Output from acceleration sensors A2 to A3, gyro sensors G2 to G3, and floor reaction force sensors F1 to F8 disposed on the sole of the vehicle.

  Next, the motion of the robot is evaluated on the motion editing system based on the sensor information acquired during the reproduction of the motion data (step S4). A method for evaluating the operation of the robot 100 based on the sensor information will be described later.

  As a result of evaluating the operation of the robot, if the predetermined evaluation criterion is not satisfied, a motion correction process is performed (step S5), and the process returns to step S2 to evaluate the motion based on the reproduction on the actual machine again. Do.

  Also, as a result of evaluating the operation of the robot, if a predetermined evaluation criterion is satisfied, a motion data file with reference data in which the sensor information obtained in step S3 is embedded as reference data is created (step S6). The entire processing routine ends.

  Here, the reference data includes sensor information, operation information, and the like that should be actually obtained during the reproduction of the motion data. Specifically, the measured value at the time of executing the motion for the angle command value for each joint, the combination of the measured value of each sensor at the time of executing the motion and the measured value after filtering the sensor output, the left and right soles The reference data includes the ZMP trajectory corrected by the stabilization control at the time of executing the motion with respect to the target ZMP trajectory at the time of editing, the measured value at the time of executing the motion with respect to the target value at the time of editing the floor reaction force sensor, and the like.

  Even if the motion of the robot described by the motion data is the same, the output of each sensor differs depending on the external environment or the work environment. For example, even if the walking pattern is the same, the output value of the sensor is different on a slope, on a gravel on which the road surface moves, or on a carpet with long hairs. By embedding the reference data in the motion data, there is an advantage that motion data corresponding to a specific working environment (or a robot use form) can be described.

  The motion data is represented, for example, as m1 (t1, p, d, g, tr) and m2 (t2, p, x, y, x, Yaw, Roll, Pitch, tr). On the other hand, reference data is, for example, S1 (t3, p, fx, fy, fz, tr), S2 (t4, p, ax, ay, az, tr), S3 (t5, p, vx, vy, tr). vz, tr). The motion data with reference data is represented, for example, by M [m1 (), m2 (), m3 ()]. However, unnecessary data can be appropriately removed. Further, the description format of motion data and its components may be appropriately changed according to the system, the purpose of the motion, and the like, and are not limited thereto.

  By evaluating and correcting the motion data according to the processing procedure shown in FIG. 5, sensor information of the real robot can be acquired through communication, and a motion can be created in consideration of the response of the real robot. Further, when the motion data created in the reference environment is executed in a different environment such as a residential environment, it can be determined whether or not the actual robot is operating as intended.

  Evaluating motion data on a real robot in this way is significant mainly in the following points.

(1) Difference in impression of acceleration:
Since the sensitivity to the speed and acceleration of each part is greatly different between the animation reproduction by computer graphics (CG) in the virtual space and the reproduction by the real machine in the real space, the adjustment is performed using a real robot. There is a need.

(2) Difference in posture impression:
Since the reproduction of animation by computer graphics in virtual space and the reproduction by real robot in real space differ greatly in the size and size of the real robot and the sensitivity of each part to the position / posture relationship, the adjustment is It is necessary to use a robot. For example, if the part used for stabilization is only the waist, the waist movement that was not noticed by computer graphics would be greatly felt by the real robot in the playback confirmation, and the desired dynamic posture could be obtained. May not be possible. Therefore, the desired dynamic posture can be corrected by increasing the stabilization region and performing partial playback while changing the priority (eg, adding the trunk and head to the stabilization region in addition to the waist). As a result, the motion of the waist of the real robot is reduced, and a desired motion can be obtained.

(3) Difficulty in estimating actuator torque (current) on actual machine:
Accurate current estimation of actual actuators requires the construction of extremely rigorous motor models and high-precision identification of each element, and simulations using those models require general computer systems such as PCs. Requires much more time than real time. For this reason, a more robust motion can be created without significantly increasing the motion creation time by performing actual measurement by reproducing the motion of the actual device and correcting the details.

(4) Difficulty in specifying accurate contact / contact time:
The deviation of the robot's point of contact with the external world and its contact timing from the planned motion results in the addition of an unknown and impulse-like external force and external force moment to the robot. The effect on the stability of the material is very large. However, construction of a strict shape model of a real robot requires high-precision identification including the external environment, and simulation of the contact state using the model requires real-time computation in a general computer system such as a PC. Requires much more time and is not realistic. Therefore, on a general PC, there is a connection between the motion generated using a simplified shape and environment model that can be processed in real time and the case where the motion is executed by a real robot in a real standard environment group. An extremely robust motion can be generated by re-executing the motion correction and the posture (motion) stabilization process using the deviation of the point and its attachment / detachment state.

  Even if the motion of the robot described by the motion data is the same, the output of each sensor differs depending on the external environment or the work environment. For example, even if the walking pattern is the same, the output value of the sensor is different on a slope, on a gravel on which the road surface moves, or on a carpet with long hairs. According to the present embodiment, there is an advantage that motion data according to a specific work environment (or a usage form of a robot) can be described by embedding reference data in motion data.

  FIGS. 6 to 9 show examples of motion data file formats. In the present embodiment, the motion data includes joint angle information, posture information, ZMP trajectory information, and sole contact information.

  As shown in FIG. 6, the joint angle information is obtained when the displacement amounts of the joint actuators constituting the degree of freedom of the joint of the legged mobile robot 100 at the time of executing the motion are arranged in time series (for each sampling time). It is series data. The record for each sampling time is composed of a combination of an angle command value for each joint and a measured value at the time of executing a motion. In the figure, R_joint name is an angle command value for the corresponding actuator at the time of editing, and M_joint name is a measured value of the actuator when the robot executes a motion.

  As shown in FIG. 7, the posture information is obtained by chronologically acquiring information of posture sensors (gyro sensors) installed at various parts on the body of the legged mobile robot 100 at the time of executing a motion (for each sampling time). ) Is time-series data. The record for each sampling time is composed of a combination of a target value at the time of motion editing for each sensor, a measured value at the time of executing the motion, and a measured value after filtering the sensor output. In the drawing, R_sensor name is a target value of the sensor at the time of editing in the corresponding sensor, M_sensor name is a measured value of the corresponding sensor at the time of executing a motion of the robot, and F_sensor name is a robot. This is a measured value after filtering the output of the corresponding sensor when the motion is executed.

  As shown in FIG. 8, the ZMP trajectory information is time-series data in which ZMP positions of the legged mobile robot 100 at the time of performing a motion are arranged in a time-series manner (for each sampling time). The record for each sampling time is composed of a combination of the target ZMP trajectory at the time of editing and the ZMP trajectory corrected by the stabilization control at the time of executing the motion for each of the left and right soles. In the figure, the R_ZMP trajectory is a target ZMP trajectory at the time of editing, and the M_ZMP trajectory is a ZMP trajectory corrected by the stabilization control at the time of executing the motion of the robot.

  As shown in FIG. 9, the sole contact information indicates that the measured values of the respective floor reaction force sensors installed on the sole of the legged mobile robot 100 at the time of executing the motion are time-series (for each sampling time). This is time-series data that is arranged. The record for each sampling time is configured by a combination of a target value at the time of editing of the floor reaction force sensor and a measured value at the time of executing a motion. In the figure, R_ground information is a target value at the time of editing, and M_ground information is a measured value at the time of executing a motion.

  FIG. 10 shows, in the form of a flowchart, a processing procedure for evaluating the motion data of the robot 100 based on the sensor information, which corresponds to step S4 of the flowchart shown in FIG.

  In step S11, the response of the actuator is evaluated. More specifically, the deviation (i.e., followability) of each of the angle command value, angular velocity command value, or angular acceleration command value for the actuator at the time of editing and each measured value when executed on an actual robot is shown. Evaluation is performed in series (step S12).

  The evaluation of the tracking performance may be performed in the frequency domain instead of the time series. When the evaluation is performed in the frequency domain, there is an advantage that a frequency-specific band can be limitedly handled, for example, by removing a natural frequency of a mechanical system.

  Note that the conversion between the time space and the frequency conversion may use the Fourier transform and the inverse Fourier transform shown in the following equation.

  As the required torque increases, the maximum speed (acceleration) of the actuator decreases, the responsiveness deteriorates, and the actuator cannot follow the target. Further, when contact with the movable angle limiting stopper or interference between the links occurs, the deviation of the high-frequency component not included in the target value increases. Therefore, the deviation between the calculated target value and the actually measured value is evaluated using the following evaluation expressions (7) to (15) (step S13).

If the above evaluation formula is not satisfied, the contents are saved (step S14), and if the evaluation is achieved, the evaluation is terminated.

In step S15, the operation status of the actuator is evaluated. More specifically, the torque value T _i (t) and the rotation speed N _i (t) of the i-th actuator when executed by the real robot are acquired in time series, and the acquired data is obtained as a characteristic of the i-th actuator. Is compared with the NT curve showing The NT curve can be obtained, for example, from the product specifications of the actuator (see FIG. 20). On the NT curve, the normal minimum torque _{Ti_min} (N) and the normal maximum torque _{Ti_max} (N) when the rotation speed is N are defined based on the NT curve.

  Then, it is evaluated whether there is any operation exceeding the limit torque of the actuator (step S16). If the evaluation has not been achieved, the content is saved (step S17), and if the evaluation has been achieved, the evaluation ends. Although the calculation of the torque value by the simulation takes a very long time and includes an error, accurate information can be obtained immediately by obtaining the torque value from the actual robot connected to the editing system.

In step S18, the posture is evaluated. First, a deviation between a posture sensor value or a ZMP trajectory planned at the time of motion editing and a sensor value or a ZMP trajectory when executed by a real robot is obtained (step S19). More specifically, the target posture on the time axis of the coordinate system origin (O'-X'Y'Z ') (for example, the local coordinate origin set at the ground contact site such as the sole) planned at the time of motion editing A _df (t) or target attitude A _df (ω) on the frequency axis

And a target posture A _dh (t) on the time axis or a target on the frequency axis of the origin (O "-X" Y "Z") of the coordinate to be controlled by the movable part (for example, a direct control target point such as the waist) Posture A _dh (ω),

And ZMP _d (t) = [ZMP _dx , ZMP _dy , ZMP _dz , t] as the target ZMP trajectory or ZMP _d (ω) = [ZMP _dx , ZMP _dy , ZMP _dz , ω] on the target ZMP trajectory the translational force and F _d as the yaw axis moment _{(t) = [F dx,} F dy, F dz, t] or _{F d (ω) = [F} dx, F dy, F dz, ω], in the actual robot base control target coordinate origin (O'-X'Y'Z ') orientation sensor measurements in the time domain in a _{mf (k) (t)} or attitude sensor measurement a _mf on the frequency axis _(k when running ₎ (Ω)

And the attitude sensor measured value A _{mh (k)} (t) on the time axis or the attitude sensor on the frequency axis at the coordinate origin (O "-X" Y "Z") of the movable part control target when executed by the real robot. Measured value A _{mh (k)} (ω)

If, ZMP _m as measured ZMP trajectory (k) (t) = [ ZMP mx, ZMP my, ZMP mz, t] or _{ZMP m (k) (ω)} = [ZMP mx, ZMP my, ZMP mz, ω] , translational force on the measurement ZMP trajectory and F _m as the yaw axis moment (k) (t) = [ F mx, F my, F mz, k, t] or _{F m (k) (ω)} = [F _{mx, F my, F mz,} k, ω], M m (k) (t) = [M mx, M my, M mz, k, t], M m (k) (ω) = [M mx, M _my , M _mz , k, ω] (A _{ef (k)} (t), A _{eh (k)} (t), ZMP _{e (k)} (t), F _{e (k)} (t), _{M e (k) (t)} , A ef (k) (ω), A eh (k) (ω), ZMP e (k) (ω), F e (k) (ω), M e (k) (Ω)) is calculated. Please refer to FIGS. 14 to 16 for the method of setting the coordinate system and the like.

  Then, the deviation between the calculated target value and the actually measured value is evaluated using the following evaluation expressions (27) to (41) (step S20).

  If the above evaluation formula is not satisfied, the contents are saved (step S21), and if the evaluation is achieved, the evaluation is terminated. The stored deviation information is used to correct the input value of the posture stabilization control block at the time of re-editing.

In step S22, the contact position is evaluated. In the present embodiment, as shown in FIG. 3, there are provided load cells F1 to F8 and infrared distance measuring sensors D1 to D8 disposed on the right and left soles, respectively. Then, the ground / offset information C _{d (i)} (t) or C _{d (i)} (ω) in each part of the sole at the time of motion editing and the load cells F1 to F8 when the motion is executed on the actual robot. And a deviation C _e from the grounding / offset information C _{m (i, k)} (t) or C _{m (i, k)} (ω) obtained from the infrared distance measuring sensors D1 to D8 (see FIG. 17). _{(i, k)} (t) = C _{d (i)} (t) −C _{m (i, k)} (t) or C _{e (i, k)} (ω) = C _{d (i)} (ω) −C _{m (i, k)} (ω) is calculated (step S23). Then, the calculated deviation is evaluated using the following equation (37) or the like (step S24). Here, n is the number of steps, and Ce _(i) is an allowable value of the grounding / leaving information.

As a method for acquiring the grounding / offset information of the grounding position, analog information F _{imz (k)} (t) output from the load cells F1 to F8 provided at each corner of the sole-grounded portion and the infrared distance measuring sensors D1 to _D8. If is less than the predetermined threshold value gamma _i the Hanarechi _{Cm (i, k) = 1} , it is determined if more than the threshold value gamma _i grounded _{C m (i, k) =} 0 and. Here, i is the identification number of the sensor attached to the sole. Alternatively, a micro switch may be provided at each corner of the sole contact portion, and the contact / ground information may be obtained based on the ON / OFF information.

  If the evaluation of the contact position has not been achieved, the content is saved (step S25), and if the evaluation has been achieved, the evaluation ends. The deviation of the robot's point of contact with the external world and its contact timing from the planned motion results in the addition of an unknown and impulse-like external force and external force moment to the robot. The effect on the stability of the material is very large. Therefore, the motion correction and the posture (motion) stabilization processing are performed again using the difference between the contact point and the contact point between the motion generated in the ideal virtual space and the case where the motion is executed by the real robot. This makes it possible to generate extremely robust motion.

In step S26, the degree of achievement of correction is evaluated. That is, the degree of improvement of the measured value is calculated for the motion corrected by the previous evaluation (step S27). For example, in the case of the actuator angle θ, if the deviation between the target value and the actually measured value at the time of the k-th measurement is θ _{e (k)} (= θ _d −θ _{m (k)} ), the degree of improvement P is k It can be expressed by the change of the deviation between the first and the (k-1) th time, that is, P = θe _(k) −θe _(k-1) . Then, the comprehensive evaluation is performed again according to the priority of each item (step S28). If the evaluation has not been achieved, the contents are saved (step S29), and if the evaluation has been achieved, the evaluation ends.

In step S30, an impact due to external contact is evaluated. More specifically, the influence of the actuator torque, the ZMP deviation, and the acceleration due to the impact is calculated (step S31), and the influence is evaluated (step S32). For example, the influence of the i-th actuator torque or ZMP deviation / acceleration due to the impact is calculated using T _i (ω) obtained by converting T _i (N) into the frequency domain or the above equations (17) to (26). I do. Then, the influence is evaluated using the above-described evaluation expressions (27) to (41).

  If the evaluation of the impact due to external contact has not been achieved, the contents are saved (step S33), and if the evaluation has been achieved, the evaluation ends. It is extremely difficult to rigorously model the outside world of a robot, especially considering the living environment of humans. Therefore, using a simple collision model that does not increase the execution time of the simulator compared to the real time, first, a motion involving a collision with the outside world is generated in an ideal virtual space, By using the deviation of the force information due to external contact with the case of executing the motion, the motion is corrected and the posture (motion) stabilization process is performed again, so that an extremely robust motion can be generated in a short time. .

  FIG. 11 shows, in the form of a flowchart, a procedure of a motion correction process corresponding to step S5 of the flowchart shown in FIG.

  Motion correction can be performed manually or automatically on the motion editing system.

  In the case of the manual correction, the correction is made while visually confirming the difference value between the command value and the measured value and the details of the improvement made by the previous correction using a graph or the like (step S41).

  On the other hand, in the case of automatic correction, the contents of the response evaluation result in step S11 are used to indicate the angle or the angular velocity or the angular acceleration to the actuator.

Is corrected (step S42). For the correction, for example, an angle calculated by the following equations (38) to (43), or an instruction value of angular velocity or angular acceleration is used. Here, Λ is a filter, and α is a gain.

  In addition, the control parameters of the actuator (such as the PID and the viscosity of the motor) are corrected using the response evaluation result, particularly the evaluation result in the frequency range (step S43).

  Specifically, the real robot performs a switching operation between the closed link state and the open link state with respect to the outside world and the work object such as bipedal walking and stable double-armed work when executing a motion, but in the high frequency region of the open link. If the response is insufficient, the D (differential) gain is increased and the set value of the viscosity is increased. When the response of the open link in the high frequency region is equal to or higher than the target value, the gains of P (proportional) and D (differential) are reduced and the set value of the viscosity is increased. On the other hand, if the response of the open link in the low frequency region is insufficient, the gains of P (proportional) and I (integral) are increased. If the response in the low frequency region of the open link is equal to or higher than the target value, the gain of I is reduced.

  If the response of the closed link in the low frequency region is insufficient, the gains of P (proportional) and I (integral) are increased. Conversely, if it is equal to or greater than the target value, the gain of I is reduced. If the response in the high frequency region of the closed link is equal to or higher than the target value, the gain of D (differential) is reduced and the set value of the viscosity is increased. Conversely, if the response is insufficient, the gain of D is increased. And lower the set value of viscosity to improve the profiling performance on walking road surfaces.

Further, the stabilization priority part of the posture stabilization processing block is changed using the contents of the evaluation result of the operation status of the actuator in step S15 (step S44). Specifically, the real robots are grouped for each part such as the lower limb, the trunk, the waist, the right upper limb, etc., but the target moment for the part group i driven by the actuator that does not satisfy the above-mentioned evaluation formula (16) is used. and smaller reconfigure compensation amount M _i, to reduce the momentum to be used for stabilization, reduce the required torque.

The content of the posture stabilization processing block is changed using the posture stability evaluation result in step S18 (step S45). Specifically, reconfigure smaller target moment compensation amount M _i for site groups i that is driven by an actuator which does not satisfy the above evaluation formula (27) - (41), to reduce the momentum to be used for stabilizing Reduce the required torque.

Further, the content of the posture stabilization processing block is changed using the evaluation result of the contact position in step S22 (step S46). Specifically, the ground / offset information C _{d (i)} (t) or C _{d (i)} (ω) at the time of motion editing and the ground / offset information obtained when a motion is executed on a real robot Deviation from C _{m (i, k)} (t) or C _{m (i, k)} (ω) Ce _{(i, k)} (ω) = C _{d (i)} (ω) −C _{m (i, k )} (Ω), using Ci _{(i, k)} (t) or Ci _{(i, k)} (ω) calculated by the following evaluation formula (44) _, or using the measured value C _{m (i , k)} (t) is used directly to newly reset the support polygon, and then the target ZMP trajectory is reset to regenerate the whole body motion.

Further, the control contents of the posture stabilization processing block in consideration of the external contact are changed using the evaluation result of the impact by the external contact in step S30 (step S47). Specifically, the whole body motion is regenerated using the input ZMP trajectory ZMP _{i (k)} (t) calculated using the following equation (45).

  Here, Λ represents, for example, a low-pass filter when improvement cannot be expected from a certain high frequency component or higher due to the frequency characteristics of the actuator, and a high-pass filter when no active improvement from a certain low frequency component or lower is not performed. A filter is used, a bandpass filter is used when both effects are expected, and a band elimination filter or a notch filter is used when it is desired to suppress the excitation of the natural frequency of the airframe. In the above equations (43) and (44), the filters for the frequency are used. However, the filters for the time domain may be used together.

  The motion editing process according to the present embodiment can reproduce and evaluate motion data on the actual robot device, correct the motion based on the sensor information acquired during the reproduction of the motion data, Another feature is that motion data in a form in which reference sensor information is embedded can be obtained as a processing result.

  Here, when the motion data is reproduced on the actual device, only an arbitrary range of the motion data is extracted and reproduced on the actual device, thereby improving the efficiency of the motion editing work.

  A motion is composed of a chronological combination of two or more poses. When a dynamic motion (continuous dynamic motion) using acceleration positively and continuously is executed, the motion is executed from the middle. However, it is impossible to make an evaluation.

  Therefore, in the present embodiment, such continuous dynamic motion can be reproduced / stopped from an arbitrary time, so that the time required for motion evaluation is extremely reduced.

  FIG. 12 shows, in the form of a flowchart, a procedure for processing reproduction / stop of a continuous dynamic motion described by motion data of the robot 100 from an arbitrary time. In the example shown in the figure, the reproduction / stop of motion data can be performed online, that is, in real time.

  First, a start time in the motion data is set by a user input or the like (step S51).

  Next, the dynamic posture at this start time is calculated (step S52). Then, a transient motion having the dynamic posture at the start time as the end point is generated (step S53), and the motion is reproduced on the actual device using the transient motion (step S54).

  Next, a stop time in the motion data is set by a user input or the like (step S55). Then, the dynamic posture at the stop time is calculated (step S56), a transient motion starting from the stop posture is generated (step S57), and the operation of the real machine is stopped using the transient motion (step S58). .

  It is not always necessary to play and stop motion data online. FIG. 18 shows a procedure for processing the reproduction / stop of the continuous dynamic motion described by the motion data of the robot 100 from an arbitrary time off-line.

  First, a start time in the motion data is set by a user input or the like (step S71). Next, a stop time in the motion data is set by a user input or the like (step S72).

  Next, the dynamic posture at this start time is calculated (step S73). Then, the dynamic posture at the stop time is calculated (step S74). Then, a transient motion having the dynamic posture at the start time as the end point is generated (step S75).

  Next, a transient motion starting from the stop posture is generated (step S76). Then, using this transient motion, the motion is reproduced on the actual device (step S77), and the operation of the actual device is stopped (step S78).

  FIG. 5 schematically shows the procedure of the motion editing processing, and FIG. 13 shows this processing procedure in more detail.

  First, motion data is created on the motion creation / editing system (step S61). For example, a user edits motion data for the legged mobile robot 100 offline on a motion editing system such as a personal computer in which a motion editing application is installed. The motion data can be edited by, for example, combining two or more poses (postures) of the robot 100 in chronological order. The pose of the robot 100 is described by the displacement amount of each joint angle. The motion data is composed of the displacement amount of each joint angle, its speed, acceleration, and the like.

  Next, the created motion data is installed in the legged mobile robot 100, and the operation is confirmed using an actual machine. However, instead of installing the motion data on the actual device, the data may be distributed to the actual device in a streaming format.

  In order to confirm this operation, the motion is stabilized by selecting a stabilizing part in the virtual space and designating its priority (step S63). Next, the operation phenomenon of the stabilized motion is confirmed on the motion creation / editing system (step S64). If the motion is a desired motion in the virtual space, the process proceeds to step S65, and the operation proceeds to the operation check using the actual device.

  If it is confirmed that the motion is not the desired motion in the virtual space, the process proceeds to S62, where manual or automatic motion editing is performed. In the motion editing, as described with reference to the flowchart of FIG. 11, correction of an angle instruction value and a control parameter to a corresponding actuator, change of a stabilization priority portion based on torque evaluation, posture evaluation or grounding position evaluation The contents of the posture stabilization processing block are changed, and the control parameters of the actuator are changed based on the response evaluation (steps S42 to S47).

  In step S65, the editing motion data is taken into the real robot. In step S66, the operation phenomenon is confirmed using the real robot.

  At this time, the motion data itself is time-series data including continuation of motion, that is, displacement and speed of each joint angle, and is sometimes long. In the present embodiment, it is possible to improve the efficiency of the motion editing work by extracting only an arbitrary range of motion data (see FIGS. 12 and 18) and reproducing it on an actual device.

  As a result of checking the motion phenomenon using the real robot, if it is found that the motion is not the desired motion, the process returns to step S62 for reselecting a stabilizing portion, redesignating its priority, and reediting the motion. . In step S62, as described with reference to the flowchart of FIG. 11, correction of the angle instruction value or control parameter to the corresponding actuator, change of the stabilization priority part based on torque evaluation, posture evaluation or ground position evaluation is performed. The contents of the posture stabilization processing block are changed, and the control parameters of the actuator are changed based on the response evaluation (steps S42 to S47).

  On the other hand, as a result of checking the operation phenomenon using the real robot, if the motion is a desired motion, the process proceeds to step S67, and the sensor information of the real robot is transferred to the motion creating / editing device. The sensor information referred to here includes, in addition to the rotation signal from the encoder provided in each joint actuator, the acceleration sensor A1 and the gyro sensor G1, which are provided in the trunk unit 40 (substantially the center of gravity of the body). Output from acceleration sensors A2 to A3, gyro sensors G2 to G3, and floor reaction force sensors F1 to F8 disposed on the sole of the vehicle.

  Next, the operation of the robot is evaluated on the motion creation / editing device based on the acquired sensor information (step S68). The evaluation of the operation of the robot is performed in accordance with the processing procedure shown in FIG. 10 described above. Is evaluated.

  Here, as a result of evaluating the operation of the robot, if the predetermined evaluation criterion is not satisfied, the process proceeds to step S70, and the evaluation content is stored. Then, after performing the motion correction processing (step S562), the operation evaluation based on the reproduction on the actual device is performed again.

  In addition, as a result of evaluating the operation of the robot, if a predetermined evaluation criterion is satisfied, the process proceeds to step S69 if the evaluation criterion is satisfied, and a final motion data file in which reference data is embedded is created (FIG. 6). To FIG. 9). By repeating the motion work shown in FIG. 13 a plurality of times, it is possible to acquire reference data in various external environments such as terrain and uneven terrain, and embed and save the data in one motion data file.

  Here, the reference data includes sensor information, operation information, and the like to be obtained during the actual reproduction of the motion data. Specifically, the measured value at the time of executing the motion for the angle command value for each joint, the combination of the measured value of each sensor at the time of executing the motion and the measured value after filtering the sensor output, the left and right soles The reference data includes the ZMP trajectory corrected by the stabilization control at the time of executing the motion with respect to the target ZMP trajectory at the time of editing, the measured value at the time of executing the motion with respect to the target value at the time of editing the floor reaction force sensor, and the like.

  Even if the motion of the robot described by the motion data is the same, the output of each sensor differs depending on the external environment or the work environment. For example, even if the walking pattern is the same, the output value of the sensor is different on a slope, on a gravel on which the road surface moves, or on a carpet with long hairs. In the present embodiment, since the motion data includes reference data collected when a motion is executed in various external environments, sophisticated and adaptive control according to the external environment during the motion is performed. It becomes possible.

  If the reference data included in the motion data includes various reference data in multiple external environments such as different road environments other than the standard environment, each reference data and the By comparing the deviation with the actually measured data, the current external environment can be estimated. Further, after estimating the external environment, it is possible to determine whether or not the reproduction of the motion data can be continued as it is according to the external environment.

  FIG. 19 shows, in the form of a flowchart, a procedure for reproducing motion data using reference data.

  During the motion playback, the measured value at the time of executing the motion for the angle command value for each joint, the combination of the measured value of each sensor at the time of executing the motion and the measured value after filtering the sensor output, the left and right soles As the reference data, the ZMP trajectory corrected by the stabilization control at the time of executing the motion with respect to the target ZMP trajectory at the time of editing, the measured value at the time of executing the motion with respect to the target value at the time of editing the floor reaction force sensor, etc. S81).

  Here, the reference data and the actually measured data are compared to determine whether it is possible to continue the motion reproduction (step S82).

  If it is determined that the continuation of the motion reproduction cannot be performed, a predetermined abnormal termination process is performed (step S83), and the entire motion reproduction process is abnormally terminated.

  On the other hand, when it is determined that the motion reproduction can be continued, the actual external environment is estimated by comparing the reference data with the actually measured data (step S84).

The comparison between the reference data and the actually measured data is performed based on the deviation obtained using the evaluation formula. Here, assuming that reference data for a plurality of external environments is prepared, and the reference data for the i-th environment is _Ar (i) and the measured data is _Am (i), the deviation is represented by the following equation (46). ).

  Then, the average of the deviations in the control cycle is obtained based on the following equation (47), and the environment i having the reference data that minimizes this can be estimated as the current external environment.

  When the reproduction of the motion data ends (step S85), the entire processing routine ends normally.

With the above configuration, in the present invention,
(1) The motion data of the robot is created by the editor device, and the created motion data is transferred to the actual robot.
(2) The robot executes the received motion data and acquires data of an internal state corresponding to the external environment at the time of executing the motion from each sensor.
(3) Further, the robot transmits the acquired internal state data to the editor device as reference data.
(4) The editor device stores the reference data obtained from the robot as motion data with reference data based on the created motion data. At that time, the motion data transmitted first may or may not be modified based on the execution result.
(5) Conversely, when the robot executes certain motion data, the motion data with reference data is compared with the internal state data of the robot.

  Thus, it is possible to estimate an external environment such as a road surface during the execution of the motion, and to know whether or not the motion has been reproduced. Note that the internal data referred to here is a value representing an internal state quantity (for example, a joint angle, each sensor value, or the like) of the robot that changes according to the external environment of the robot.

[Supplement]
The present invention has been described in detail with reference to the specific embodiments. However, it is obvious that those skilled in the art can modify or substitute the embodiment without departing from the scope of the present invention.

  The gist of the present invention is not necessarily limited to products called “robots”. In other words, if it is a mechanical device or other general mobile device that performs motion similar to human motion using electric or magnetic action, it is a product belonging to another industrial field such as a toy etc. Even if there is, the present invention can be similarly applied.

  In short, the present invention has been disclosed by way of example, and the contents described in this specification should not be interpreted in a limited manner. In order to determine the gist of the present invention, the claims described at the beginning should be considered.

FIG. 1 is a diagram showing an external configuration of a legged mobile robot 100 to be subjected to motion editing by the motion editing system according to the present invention. FIG. 2 is a diagram showing an external configuration of the legged mobile robot 100 to be subjected to motion editing by the motion editing system according to the present invention. FIG. 3 is a diagram schematically illustrating a degree of freedom configuration of the legged mobile robot 100 used in the embodiment of the present invention. FIG. 4 is a diagram schematically illustrating a control system configuration of the legged mobile robot 100. FIG. 5 is a diagram schematically showing a processing flow in the motion editing system according to one embodiment of the present invention. FIG. 6 is a diagram showing a data structure of joint angle information included in motion data. FIG. 7 is a diagram showing a data structure of the posture information included in the motion data. FIG. 8 is a diagram showing a data structure of ZMP trajectory information included in the motion data. FIG. 9 is a diagram showing a data structure of the sole contact information included in the motion data. FIG. 10 is a flowchart illustrating a processing procedure for evaluating the motion data of the robot 100 based on the sensor information. FIG. 11 is a flowchart showing the procedure of the motion correction process. FIG. 12 is a flowchart showing a procedure for processing a continuous dynamic motion described by the motion data of the robot 100 from a given time to playback and stop. FIG. 13 is a diagram showing the flowchart of the motion editing process shown in FIG. 5 in further detail. FIG. 14 is a diagram for explaining the posture evaluation of the real robot. FIG. 15 is a diagram for explaining the posture evaluation of the real robot. FIG. 16 is a diagram for explaining the posture evaluation of the real robot. FIG. 17 shows the ground / offset information C _{m (i, k)} (t) or C _{m (i,} obtained from the load cells F1 to F8 and the infrared distance measuring sensors D1 to D8 when a motion is executed on the real robot. _k) is a view showing (ω). FIG. 18 is a flowchart illustrating another example of the procedure for processing the reproduction / stop of the continuous dynamic motion described by the motion data of the robot 100 from an arbitrary time. FIG. 19 is a flowchart showing a procedure for reproducing motion data using reference data. FIG. 20 is a diagram showing an example of the NT curve.

Explanation of reference numerals

DESCRIPTION OF SYMBOLS 30 ... Head unit, 40 ... Trunk unit 50 ... Arm unit, 51 ... Upper arm unit 52 ... Elbow joint unit, 53 ... Forearm unit 60 ... Leg unit, 61 ... Thigh unit 62 ... Knee joint unit, 63 ... Shin unit 80: Control unit 81 ... Main control unit 82 ... Peripheral circuit 100 ... Legged mobile robot

Claims (48)

A motion editing device for editing motion data of a robot device having a plurality of joint degrees of freedom and a driving unit and a sensor for driving the joint,
Motion data creating means for creating motion data of the robot device;
Output means for outputting the motion data to the robot device;
Reference data obtaining means for obtaining, as reference data, information composed of sensor information or motion information of the robot device when the output motion data is executed by the robot device,
Based on the motion data and the reference data created by the motion data creating means, motion data with reference data creating means to create motion data with reference data,
A motion editing device comprising: Further comprising motion data evaluation means for evaluating the created motion data based on the acquired reference data;
The motion editing apparatus according to claim 1, wherein The apparatus further comprises a motion data correction unit that corrects the motion data based on an evaluation result by the motion data evaluation unit,
The motion editing apparatus according to claim 2, wherein The motion data with reference data creating means creates motion data with reference data obtained by adding the reference data to motion data satisfying an evaluation criterion in the motion data evaluation means,
The motion editing device according to claim 3, wherein The output means outputs the motion data to the robot device in a streaming format,
The motion editing apparatus according to claim 1, wherein The motion data with reference data creating means creates motion data with reference data including joint angle information including a combination of an angle command value for each joint and a measurement value at the time of executing the motion in the reference data.
The motion editing apparatus according to claim 1, wherein The motion data with reference data creating means is configured to combine a target value at the time of creating a motion for each sensor, a sensor output at the time of executing the motion, and a measured value obtained by filtering the sensor output. Creates motion data with reference data included in the reference data,
The motion editing apparatus according to claim 1, wherein The motion data creating means with reference data refers to ZMP trajectory information composed of a combination of a target ZMP trajectory for editing of each of the left and right soles and a ZMP trajectory corrected by the stabilization control during motion execution. Create as motion data with data,
The motion editing apparatus according to claim 1, wherein The motion data creating means with reference data is a sole contact information or a contact formed by a combination of a target value of the floor reaction force sensor at the time of creating the motion data and a measurement value of the floor reaction force sensor at the time of executing the motion. Create information as motion data with reference data,
The motion editing apparatus according to claim 1, wherein The motion data evaluation means evaluates the motion data based on a time series or a frequency domain when the robot device executes the motion data,
The motion editing apparatus according to claim 2, wherein The motion data evaluation means obtains an actuator torque value and a rotation speed in a time series when the motion data is executed by the robot device, and compares the obtained data with a characteristic of an actuator represented by an NT curve. Evaluate motion data based on
The motion editing apparatus according to claim 2, wherein The motion data evaluation means calculates a difference between a posture sensor value or a ZMP trajectory planned at the time of motion data creation by the motion data creation means and a sensor value or a ZMP trajectory when executed by a real robot. Evaluate motion data based on device attitude,
The motion editing apparatus according to claim 2, wherein The motion data evaluation means calculates a difference between the posture of the robot device when creating the motion data by the motion data creation means and a measurement value when executing the motion data with the robot device, Performing motion data evaluation based on grounding or contact of the robotic device;
The motion editing apparatus according to claim 2, wherein The motion data evaluation means is an evaluation result when executing the motion data corrected by the motion data correction means based on the previous evaluation, and the motion data correction means based on the current evaluation. By comparing the evaluation results when the corrected motion data is executed, the degree of correction achieved by correcting the motion data is evaluated.
The motion editing device according to claim 3, wherein The motion data correction unit corrects an angle instruction value to the actuator based on the actuator responsiveness evaluation result, or corrects a control parameter of the actuator,
The motion editing device according to claim 3, wherein The motion data correction unit corrects motion data by a posture stabilization process based on an evaluation result of the actuator torque.
The motion editing device according to claim 3, wherein The motion data correction means corrects the motion data by the posture stabilization process based on the evaluation result of the contact with the ground or the contact.
The motion editing device according to claim 3, wherein The motion data correction unit corrects the motion data by a posture stabilization process in consideration of the external contact, based on the evaluation result of the impact due to the external contact.
The motion editing device according to claim 3, wherein A motion editing apparatus for editing motion data of a robot apparatus having a plurality of joint degrees of freedom and a driving unit for driving the joint,
Motion start time setting means for setting the start time of the motion in the motion data,
Start transient motion creating means for creating a transient motion with the posture of the robot device at the start time as the end point of the motion,
Motion stop time setting means for setting a stop time of the motion in the motion data,
Stop transient motion creating means for creating a transient motion with the posture of the robot device at the stop time as a start point of the motion,
A motion editing device comprising: In a robot apparatus having a plurality of joint degrees of freedom and a driving unit for driving the joint,
Acquisition means for acquiring motion data with reference data created by the motion editing apparatus according to claim 1;
Motion data execution means for executing the acquired motion data with reference data,
A robot device comprising: Sensor information acquisition means for acquiring sensor information or operation information of the robot device when executing the motion data by the motion data execution means,
Output means for outputting information comprising sensor information or operation information to the motion editing apparatus according to claim 1 as reference data;
The robot apparatus according to claim 20, further comprising: In a robot device having a sensor and a plurality of joint degrees of freedom and driving means for driving the joint,
Motion data acquisition means for acquiring motion data with reference data including actual measurement data of the sensor when the robot device executes the motion data as reference data,
Motion data execution means for executing the motion data with reference data,
Measurement data acquisition means for acquiring measurement data of the sensor when executing the motion data with the reference data,
External environment estimating means for estimating the current external environment by comparing the reference data and the measured data,
A robot device comprising: In a robot apparatus having a sensor and a plurality of joint degrees of freedom and driving means for driving the joint,
Motion data acquisition means for acquiring motion data with reference data including actual measurement data of the sensor when the robot device executes the motion data as reference data,
Motion data execution means for executing the motion data with reference data,
Means for acquiring the measured data of the sensor when executing the motion data with the reference data,
By comparing the reference data and the measured data, determining means for determining whether or not execution of the motion data can be continued,
Terminating means for terminating the execution of the motion data in response to the determination means determining that the execution of the motion data cannot be continued;
A robot device comprising: In a motion editing system for creating motion data of a robot apparatus having a plurality of joint degrees of freedom and a driving unit for driving the joint,
Motion data creating means for creating motion data of the robot device;
Output means for outputting the motion data;
A robot device for acquiring and executing the output motion data,
Reference data obtaining means for obtaining, as reference data, information configured from sensor information or operation information of the robot device when the output of the motion data is executed by the robot device,
Based on the motion data and the reference data created by the motion data creating means, motion data with reference data creating means to create motion data with reference data,
A motion editing system, comprising: Motion data of the robot device,
Reference data composed of sensor information or motion information obtained from the robot device when the robot device executes the motion data,
Is stored as motion data with reference data,
A storage medium characterized by the above-mentioned. A motion editing method for editing motion data of a robot device having a plurality of joint degrees of freedom and a driving unit and a sensor for driving the joint,
A motion data creation step of creating motion data of the robot device;
An output step of outputting the motion data to the robot device;
A reference data acquisition step of acquiring information composed of sensor information or operation information of the robot device when the output motion data is executed by the robot device as reference data,
Based on the motion data and the reference data created by the motion data creation step, based on the reference data, motion data with reference data to create motion data with reference data,
A motion editing method, comprising: A motion data evaluation step of evaluating the created motion data based on the acquired reference data;
The motion editing method according to claim 26, wherein: Further comprising a motion data correction step of correcting the motion data based on the evaluation result by the motion data evaluation step,
28. The motion editing method according to claim 27, wherein: In the motion data creation step with reference data, in the motion data evaluation step to create motion data with reference data obtained by adding the reference data to motion data that satisfies the evaluation criteria,
29. The motion editing method according to claim 28, wherein: In the output step, the motion data is output to the robot device in a streaming format.
The motion editing method according to claim 26, wherein: In the motion data creation step with reference data, to create motion data with reference data including joint angle information composed of a combination of an angle command value for each joint and a measurement value at the time of the motion is included in the reference data,
The motion editing method according to claim 26, wherein: In the motion data creating step with reference data, posture information including a combination of a target value at the time of creating a motion for each sensor, a sensor output at the time of executing the motion, and a measured value obtained by filtering the sensor output. Creates motion data with reference data included in the reference data,
The motion editing method according to claim 26, wherein: In the motion data creation step with reference data, the ZMP trajectory information composed of a combination of the target ZMP trajectory at the time of editing for each of the left and right soles and the ZMP trajectory corrected by the stabilization control at the time of executing the motion is referred to. Create as motion data with data,
The motion editing method according to claim 26, wherein: In the motion data creating step with reference data, the sole contact information or contact information composed of a combination of a target value of the floor reaction force sensor at the time of creating the motion data and a measurement value of the floor reaction force sensor at the time of executing the motion. Create information as motion data with reference data,
The motion editing method according to claim 26, wherein: In the motion data evaluation step, to evaluate the motion data based on a time series or frequency domain, the follow-up performance when executing the motion data in the robot device,
28. The motion editing method according to claim 27, wherein: In the motion data evaluation step, the actuator torque value and the rotation speed when the robot device executes the motion data are obtained in a time series, and the obtained data is compared with the characteristics of the actuator represented by the NT curve. Evaluate motion data based on
28. The motion editing method according to claim 27, wherein: In the motion data evaluation step, a difference between a posture sensor value or a ZMP trajectory planned at the time of creating motion data in the motion data creation step and a sensor value or a ZMP trajectory when executed by a real robot is calculated. Evaluate motion data based on device attitude,
28. The motion editing method according to claim 27, wherein: In the motion data evaluation step, the difference between the posture of the robot device at the time of creating the motion data in the motion data creation step and a measurement value when executing the motion data in the robot device is calculated. Performing motion data evaluation based on grounding or contact of the robotic device;
28. The motion editing method according to claim 27, wherein: In the motion data evaluation step, the evaluation result when executing the motion data corrected by the motion data correction means based on the evaluation before the previous time, and the motion data correction means based on the current evaluation By comparing the evaluation results when the corrected motion data is executed, the degree of correction achieved by correcting the motion data is evaluated.
29. The motion editing method according to claim 28, wherein: In the motion data correction step, based on the actuator response evaluation result, to correct the angle instruction value to the actuator, or to correct the control parameters of the actuator,
29. The motion editing method according to claim 28, wherein: In the motion data correction step, based on the evaluation result of the actuator torque, to correct the motion data by posture stabilization processing,
29. The motion editing method according to claim 28, wherein: In the motion data correction step, based on the evaluation result of the touchdown or contact, to correct the motion data by posture stabilization processing,
29. The motion editing method according to claim 28, wherein: In the motion data correction step, based on the evaluation result of the impact due to the external contact, the motion data is corrected by a posture stabilization process in consideration of the external contact,
29. The motion editing method according to claim 28, wherein: In a motion editing method for editing motion data of a robot device having a plurality of joint degrees of freedom and a driving unit for driving the joint,
A motion start time setting step of setting a motion start time in the motion data;
A starting transient motion creating step of creating a transient motion with the posture of the robot device at the start time as an end point of the motion,
A motion stop time setting step of setting a motion stop time in the motion data;
A stop transient motion creating step of creating a transient motion with the posture of the robot device at the stop time as a start point of the motion,
A motion editing method, comprising: A method of controlling a robot apparatus having a plurality of joint degrees of freedom and a driving unit for driving the joint,
An acquiring step of acquiring motion data with reference data created by the motion editing apparatus according to claim 1;
A motion data execution step of executing the acquired motion data with reference data,
A control method for a robot device, comprising: A sensor information acquisition step of acquiring sensor information or operation information of the robot device when executing the motion data in the motion data execution step,
An output step of outputting information composed of sensor information or operation information to the motion editing apparatus according to claim 1 as reference data;
The control method for a robot device according to claim 45, further comprising: In a control method of a robot apparatus having a sensor and a plurality of joint degrees of freedom and a driving unit for driving the joint,
A motion data acquisition step of acquiring motion data with reference data including actual measurement data of a sensor when the robot device executes the motion data as reference data,
A motion data execution step of executing the motion data with the reference data,
An actual measurement data acquisition step of acquiring actual measurement data of the sensor when executing the motion data with the reference data,
By comparing the reference data and the actual measurement data, an external environment estimation step of estimating the current external environment,
A control method for a robot device, comprising: A control method for a robot apparatus having a sensor and a plurality of joint degrees of freedom and a driving unit for driving the joint;
A motion data acquisition step of acquiring motion data with reference data including actual measurement data of a sensor when the robot device executes the motion data as reference data,
A motion data execution step of executing the motion data with the reference data,
Acquiring actual measurement data of the sensor when executing the motion data with the reference data,
A comparing step of comparing the reference data and the actually measured data to determine whether or not execution of the motion data can be continued;
An end step of terminating the execution of the motion data in response to determining that the execution of the motion data cannot be continued in the determining step;
A control method for a robot device, comprising:

Images