WO2005046942A1 - ロボットの駆動方法 - Google Patents
ロボットの駆動方法 Download PDFInfo
- Publication number
- WO2005046942A1 WO2005046942A1 PCT/JP2004/016968 JP2004016968W WO2005046942A1 WO 2005046942 A1 WO2005046942 A1 WO 2005046942A1 JP 2004016968 W JP2004016968 W JP 2004016968W WO 2005046942 A1 WO2005046942 A1 WO 2005046942A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image data
- robot
- operation command
- image
- moving object
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims description 92
- 238000005516 engineering process Methods 0.000 claims description 19
- 238000000513 principal component analysis Methods 0.000 claims description 8
- 230000000875 corresponding effect Effects 0.000 abstract 2
- 230000002596 correlated effect Effects 0.000 abstract 1
- 210000003811 finger Anatomy 0.000 description 40
- 238000004364 calculation method Methods 0.000 description 23
- 238000010586 diagram Methods 0.000 description 20
- 238000012545 processing Methods 0.000 description 19
- 238000000605 extraction Methods 0.000 description 16
- 210000003813 thumb Anatomy 0.000 description 16
- 210000001503 joint Anatomy 0.000 description 14
- 238000005311 autocorrelation function Methods 0.000 description 12
- 238000004422 calculation algorithm Methods 0.000 description 12
- 239000011159 matrix material Substances 0.000 description 12
- 230000008569 process Effects 0.000 description 12
- 238000000611 regression analysis Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 6
- 210000004247 hand Anatomy 0.000 description 6
- 238000005070 sampling Methods 0.000 description 6
- 230000007547 defect Effects 0.000 description 5
- 230000009467 reduction Effects 0.000 description 5
- 238000006243 chemical reaction Methods 0.000 description 4
- 230000001186 cumulative effect Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000005484 gravity Effects 0.000 description 4
- 238000005314 correlation function Methods 0.000 description 3
- 238000000354 decomposition reaction Methods 0.000 description 3
- 238000006073 displacement reaction Methods 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 210000004932 little finger Anatomy 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 241000282412 Homo Species 0.000 description 2
- 238000005452 bending Methods 0.000 description 2
- 239000011248 coating agent Substances 0.000 description 2
- 238000000576 coating method Methods 0.000 description 2
- 238000013500 data storage Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000008707 rearrangement Effects 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 241001465754 Metazoa Species 0.000 description 1
- 206010049816 Muscle tightness Diseases 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 210000000988 bone and bone Anatomy 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000005057 finger movement Effects 0.000 description 1
- 230000007274 generation of a signal involved in cell-cell signaling Effects 0.000 description 1
- 210000002478 hand joint Anatomy 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000003278 mimic effect Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000001356 surgical procedure Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 210000001364 upper extremity Anatomy 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000021542 voluntary musculoskeletal movement Effects 0.000 description 1
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J15/00—Gripping heads and other end effectors
- B25J15/0009—Gripping heads and other end effectors comprising multi-articulated fingers, e.g. resembling a human hand
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1694—Programme controls characterised by use of sensors other than normal servo-feedback from position, speed or acceleration sensors, perception control, multi-sensor controlled systems, sensor fusion
- B25J9/1697—Vision controlled systems
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J13/00—Controls for manipulators
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/42—Recording and playback systems, i.e. in which the programme is recorded from a cycle of operations, e.g. the cycle of operations being manually controlled, after which this record is played back on the same machine
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
Definitions
- the present invention relates to a method for driving a robot, and more particularly, to causing a moving object or a simulated object corresponding to the robot to perform a desired operation in order to operate the robot, and an image of the moving object or the simulated object.
- the present invention relates to a method of driving a robot suitable for causing the robot to perform the same operation as data.
- technologies related to robots that can reproduce the movement of a person by imitating the movement of a person include, for example, when an operator walks with a sensor attached, or when the operator puts a data glove equipped with a plurality of sensors on the hand, Then, the robot is driven based on the information output from these sensors.
- Non Patent Literature 1 and Non Patent Literature 2 disclose this type of technology
- Non-Patent Document 2 A paper entitled “Imitation and primitive symbol acquisition of humanoids by the integrated mimesis loop," by T. Inamura, Y. Nakamura, H. Ezaki and I. Toshima: Pro c. Of Intl. Conf on Robotics and Automation (ICRA 2001), p. 4208—p. 4213, published in 2001
- Another object of the present invention is to make it possible to drive a robot stably and at high speed even when the shape of a moving object or a simulated object is complicated or self-occluding on an image frequently occurs. To provide a method for driving a robot.
- Another object of the present invention is to provide a driving method of a robot that can increase the accuracy of specifying input motion image data.
- Another object of the present invention is to provide a driving method of a robot capable of quickly collating input motion image data with a large number of image data stored in advance.
- Still another object of the present invention is to provide a method of driving a robot by eliminating defect setting properties by imitation learning.
- the robot driving method of the present invention drives the robot in accordance with the following steps in accordance with the operation command.
- a plurality of image data of the moving object or the simulated object when the moving object or the simulated object corresponding to the robot performs the same operation as the predetermined operation, and the operation command corresponding to the plurality of image data are obtained.
- a pre-collection operation command collected in advance is stored in the image-corresponding operation command storage means in association with the command.
- the method for collecting the pre-collection operation command in advance is optional.
- the moving object or its simulated object is caused to perform a desired operation, and at this time, the operation image data of the moving object or its simulated object is acquired in time series as robot operation image data.
- the image data corresponding to the motion image data included in the robot motion image data is stored in the image-corresponding motion command storage means, and the image data is specified in a time series from a plurality of image data. Is given to the robot as an operation command.
- image correspondence The image data stored in the operation command storage means may be data accompanied by feature data indicating characteristics of the image obtained by performing the processing based on the photographed image or data itself of the feature data. Of course it is good.
- the method of specifying image data corresponding to the motion image data included in the robot motion image data is arbitrary. If the image data is feature amount data, the image data corresponding to the motion image data can be specified based on the similarity of the feature amounts.
- the same operation as that of the moving object or the simulated object without using a sensor can be performed simply by obtaining an image of the moving object or the simulated object corresponding to the robot. Can be done by a robot. Therefore, according to the present invention, it is possible to easily drive a robot using image data as input data.
- the correspondence between the image data included in the robot operation image data and the plurality of image data stored in the image corresponding operation command storage means can be determined based on the similarity between the two.
- a plurality of image data for matching based on a feature amount of the operation image data is specified. It is preferable to select the image data and specify an image corresponding to the motion image data based on the similarity between the plurality of image data for collation and the motion image data.
- the robot is driven by executing the following first to fifth steps.
- a moving object corresponding to the robot, a plurality of sensors provided on the moving object for detecting the movement of the moving object, and an operation command generation for generating an operation command based on the outputs of the plurality of sensors Use the device.
- an operation command generated by the operation command generator based on the outputs of the plurality of sensors when the moving object performs a predetermined operation is stored as a pre-collection operation command.
- the moving object is typically an entire human or a moving part such as a human hand or foot, but the moving object may be an animal other than a human.
- a plurality of image data of the moving object when the moving object performs the predetermined operation in the first step is obtained, or the moving object or its simulated object performs the predetermined operation.
- Multiple image data of the moving object or its simulated object when the same operation is performed Is obtained in chronological order.
- the “moving object” in the “moving object or its simulated object” may be the same as the moving object in the first step, but has the same shape and structure (in other words, the same joint at the same place). Yes) and another object may be used as long as it can perform the same operation.
- the “simulated object” in the “moving object or its simulated object” refers to an object created by using an simulated object generation technology such as a computer graphic technology, or a doll of the operable object.
- the image data obtained in this case is pseudo image data.
- the imitation image data is computer graphic image data.
- the method or method of acquiring a plurality of image data in a time series is arbitrary, and may be an image taken by any of a monocular camera, a binocular camera, and a multi-camera.
- the moving object is a human hand
- a plurality of image data include image data in which individual differences appearing in the human hand are taken into consideration
- the differences that cause individual differences include the curvature of the base position of the four fingers other than the thumb, the degree of overhang of the base of the thumb, the degree of opening of two adjacent fingers, and the reference angle of the two opened fingers.
- Such individual differences may be caused by the basic image already created using the above-mentioned different elements as parameters when creating image data (pseudo image data) using computer graphics technology or other similar artifact generation technology. It can be easily obtained by changing the data (by changing the parameters, creating the aforementioned different elements, such as the bone length, the angle of movement t of the joints, and the like). Therefore, it is not difficult to create image data that takes into account individual differences.
- the plurality of image data created in the second step may include a plurality of resolution-changed image data created by changing the resolution of the plurality of image data. This is because changing the resolution of the image data may make the characteristics of the image more apparent. The resolution of image data can be easily changed if there is basic image data.
- the surface of the moving object including the plurality of sensors is covered with a covering including the plurality of sensors, and simultaneously with the first step, a plurality of image data of the moving object is obtained. You may make it acquire.
- the coating has a configuration capable of eliminating unnecessary information as much as possible. Therefore, it is preferable that the color should be as plain as possible (one color, no pattern: no unnecessary unevenness). From the image data of the moving object covered with such a covering, features (especially features of contours) included in the image data can be acquired in a short time and with high accuracy.
- a plurality of image data obtained in the second step are associated with each image data included in the plurality of image data obtained in the second step and the pre-collection operation command.
- the pre-collection operation command is stored in the image-corresponding operation command storage means.
- uncreated (unphotographed) image data between the previous image data and the subsequent image data acquired in the second step, which includes only the actually captured image is obtained by computer graphic technology.
- the created one can also be stored.
- the pre-collection operation command corresponding to the uncreated image data is created by estimation based on the pre-collection operation command corresponding to the previous image data and the pre-collection operation command corresponding to the subsequent image data.
- the moving object or its simulated object performs a desired operation, and at that time, the motion image data of the moving object or its simulated object is time-sequentially used for the robot operation. Acquire as image data. In this case, it is not necessary to attach a sensor to the moving object.
- the image data corresponding to the motion image data included in the robot motion image data is specified in time series from a plurality of image data (basic data) stored in the image corresponding motion command storage means. . It is preferable that the correspondence between the motion image data included in the robot motion image data and the plurality of image data stored in the image-corresponding motion command storage means is determined based on the similarity between the two. The method of obtaining the similarity is arbitrary, and is not limited to a specific method.
- the robot acquires the pre-collection operation command corresponding to the specified image data as the operation command. To give
- the hand has a high defect setting property. Therefore, when the present invention is applied to driving a robot hand, it is possible to drive the robot hand more easily than before.
- a human hand or a simulated object is used as the moving object.
- a data glove worn on a human hand is used. This data glove has a structure in which a plurality of sensors are mounted on the glove main body at positions for detecting the movement of the motion part of the human hand corresponding to the motion part of the robot hand.
- each feature amount of a plurality of image data is individually calculated (feature amount step).
- a principal component analysis is performed on each feature amount of the plurality of image data, a principal component score is calculated for each of the plurality of image data, and the first principal component power is calculated based on the cumulative contribution ratio.
- Each rearranged image data in the image data source is stored in association with the pre-collection operation command.
- a principal component score is also obtained for the input motion image data, and a plurality of k types of image data source cards are collated based on the principal component score.
- Select each image data Create multiple types of motion image data with different resolutions by changing the resolution of the motion image data, and obtain the principal component score from these multiple types of motion image data with different resolutions. Therefore, based on the principal component score, k kinds of image data source powers may be selected from a plurality of image data for comparison. A plurality of image data for this collation are selected by a predetermined number based on the principal component score for each principal component from each image data source.
- the number of image data to be compared can be reduced as compared with the case where all stored image data and the operation image data are compared, and the advantage that the matching time is reduced is obtained.
- the similarity between the motion image data and the plurality of image data for comparison may be determined based on the principal component score, and the image data corresponding to the motion image data may be specified based on the similarity.
- the first step is an operation to be performed in advance, the first step is separated, and in the driving method used for the device for driving the robot, only the second to fifth steps are used. It is natural that the robot can be driven.
- FIG. 1 is a plan view of a known multi-fingered robot hand 1 to be controlled in the present embodiment developed by the inventors.
- FIG. 2 is a diagram showing an arrangement configuration of an actuator arranged inside the robot hand of FIG. 1.
- FIG. 3 is a diagram showing a tracking performance for a sine wave input.
- FIG. 4 is a diagram showing a tracking performance for a sine FM modulated wave input.
- FIG. 5 is a diagram showing a schematic configuration of an example of a data glove used as a remote command device.
- FIG. 6 is a diagram showing a schematic configuration of an example of a control device for a multi-fingered robot hand and a device for implementing the driving method of the present invention.
- FIG. 7 is a flowchart showing an algorithm of software used when a main part of the correspondence determining means of FIG. 6 is realized using a computer.
- FIG. 8 is a diagram showing a software anorego- rism used when the first to third steps are implemented using a computer.
- FIG. 9 is a diagram used to explain first to third steps.
- FIG. 10 is a diagram showing a software algorithm used when the fourth and fifth steps are implemented using a computer.
- FIG. 11 is a diagram showing an example of a local pattern for a higher-order local autocorrelation feature.
- FIG. 12 (A) to (C) show different elements of individual differences appearing in human hands.
- FIG. 13 is a diagram showing a difference in image data when the curvature of the base position and the degree of overhang of the thumb base are changed by changing the parameters of the computer graphic editing software.
- FIG. 14 is a diagram showing three types of image data with different resolutions.
- FIG. 15 (A) and (B) are views showing a state where one screen composed of 320 ⁇ 240 pixels is divided into 4 ⁇ 4 screens and a state where it is divided into 8 ⁇ 8 screens.
- FIG. 16 is a diagram showing an image obtained by rearranging image data.
- FIG. 17 is a flowchart showing a procedure for creating a database of image data.
- FIG. 18 is a flowchart showing the flow of a specific collation procedure.
- Fig. 19 is a conceptual diagram used to explain another embodiment of the method of the present invention.
- FIG. 20 is a flowchart showing an algorithm for creating a database.
- FIG. 21 is a flowchart showing details of the algorithm in FIG. 20.
- FIG. 22 is a diagram showing image data obtained by extracting a contour.
- FIG. 23 is a diagram showing an example of a local pattern for a higher-order local autocorrelation feature used for creating a database.
- FIG. 24 is a diagram used to explain extraction of a feature amount.
- FIG. 25 is a flowchart showing an algorithm when selecting image data in which image data corresponding to operation image data is stored in a database.
- FIG. 26 is a flowchart showing details of the algorithm in FIG. 25.
- FIG. 27 is a diagram used to explain clustering.
- Ogasawara 'Hand pose estimation using multi-viewpoint silhouette images, Proc. 2001 IEEE / RSJ Intl. Conf. On Intelligent Robots and Systems (IROS'2001), pp.1989- 1996, 2001. MH Jeong, Y. Kuno, N. Shimada and Y. Shirai: "Recognition of shape-changing hand gestures, IEICE Transactions Division D, E85—D, 10, pp.1678—1687, 2002. N. Shimada, K. Kimura , Y. Kuno and Y. Shirai: "3-D hand posture estimation by indexing monocular silhouette images, Proc. 6th Workshop on Frontier of Computer Vision, pp. 150-155, 2000).
- upper limb control such as humanoid robots has at least poor setting problems such as trajectory selection, combination of joint angles, and combination of muscle tension
- Y. Uno, K. Kawato and R. Suzuki entitled Formation and control of optimal trajectory in human multi-joint arm movement—minimum torque—change model ', Ayumi: Biologic Cybernetics, 61, pp. 89-101, 1989 .;
- both the poor setting and the sexuality have been successfully solved by humans to realize complex and elaborate voluntary movements.
- Speaking of empirical power you can also say that you have exercise experience with your own body, that is, physicality (R.
- FIG. 1 is a plan view of a known multi-fingered robot hand 1 to be controlled in the present embodiment developed by the present inventors
- FIG. 2 is an actuator arranged inside the robot hand of FIG.
- FIG. 2 is a diagram showing an arrangement configuration of 2 (micro motor with built-in encoder).
- the details of the multi-finger bot hand are disclosed in Japanese Patent Application Laid-Open No. 2003-117873.
- the movable range of each joint of the multi-fingered robot hand 1 is such that the inside and outside rotation of the thumb is 120 ° and the inside and outside rotation is 60 °.
- the bending and extension of each of the five fingers are performed in conjunction with the second joint, the movable range of the second joint is 112 °, the first joint is 7Z10 times the movable range of the second joint, and the third joint is Is 5Z7 times the movable range of the second joint.
- the opening and closing of the finger is linked to the ring finger, the index finger and little finger move, and the middle finger is not powered.
- the movable range of the ring finger is 15 °
- the index finger is 13 °
- the little finger is 31 °.
- the actuator 2 of each joint uses a DC micromotor.
- Actuator 2 is installed inside the palm for opening and closing (abduction) between the second joint of each finger, the inside and outside rotation of the thumb, the inside and outside rotation of the thumb, and the four fingers other than the thumb.
- the control of each joint operates in response to a command from a control computer (not shown), and each actuator is controlled via a DZA converter and a drive unit 3 in which a dedicated driver (micro servo) for a small servo motor is arranged. This is done by changing the voltage applied to the motor.
- the motor of each actuator 2 has a built-in incremental shaft encoder. The output of this encoder can be used for position control in addition to detection of shaft speed and rotation direction.
- the angle information of each joint has a mechanism that is fed back to a control computer (not shown) via an encoder having a higher resolution through a gear.
- the resolution of the encoder itself is 16 [PulseZRevolution].
- a gear with a reduction ratio of 1Z400 is provided in the encoder of the finger opening / closing section, and the conversion resolution is 6400 [PulseZRevolution].
- Other encoders have gears with a reduction ratio of 1 Z50, and the conversion resolution is 800 [PulseZRevolution].
- FIG. 3 shows the tracking performance for sine wave input
- Fig. 4 shows the tracking performance for sine FM modulated wave input. From these results, it can be seen that a slow operating force, a relatively fast operation, and good tracking characteristics can be obtained even for a modulated wave.
- FIG. 5 is a diagram showing a schematic configuration of an example of the data glove 4 used as a remote command device.
- the data glove 4 has a configuration in which position sensors are provided at approximately 16 positions around the joint of the glove-shaped main body.
- cyberglove (trademark) manufactured by Virtual Technologies is used as a data glove.
- the sensor 5 of this data glove has one MP (metacarpal joint) and one PIP (proximal interphalangeal joint) for each finger except the thumb, and MP, PIP, and DIP (distal internodes) for the thumb. Joints), and a sensor for measuring abduction between each finger and a sensor at the center of the palm.
- Each of the sensors 5 installed at a total of 16 locations has a different degree of distortion (strain). It has a specification to output at 30-40Hz.
- the sensor 5 installed at the center of the palm measures the degree of distortion of the entire palm, and the humanoid robot hand does not have such a mechanism. Is not used.
- the finger glove movement of a person is converted into data by a data glove worn by the operator, taken into a computer, and converted into a control command value of a multi-fingered robot hand.
- the same operation as that of the operator's hand can be realized by the robot hand.
- one simple method is to convert data output from a cyber glove into a three-dimensional coordinate position and to obtain angle information of a part that is paired with a joint of the robot hand. There is a means to take out in series, execute it by the robot hand, and operate the robot hand.
- the 16 values output from each sensor in the data glove are converted into the X, Y, and Z coordinates, pitch angle, single angle, and roll angle of each joint, and each unit time As many as 100 data are generated, and it is not optimal to extract and use the data of eight joints necessary for the operation of the robot node from the viewpoint of calculation efficiency and time efficiency.
- a multiple regression analysis is performed on the output of the data groove 4 measured in this way and the time series joint angle pattern of the robot and the robot used for the measurement to derive a partial regression coefficient for each joint.
- the following equation is a multiple regression equation used in the present embodiment.
- the objective variable ⁇ i is a command value for each joint to be controlled by the robot hand 1
- an explanatory variable X is an output value of the data glove 4
- a is a partial regression coefficient
- a is a partial regression coefficient.
- the residual Since the value obtained by the multiple regression equation of min On is the value of the joint angle of each finger in the data glove 4, it is sufficient to control this value as the angle command value of the robot node.
- FIG. 6 shows a schematic configuration of an example of a configuration of a control device for a multi-fingered robot hand used to obtain a pre-collection operation command and a device for driving a robot hand using the robot driving method of the present invention.
- the control device used to collect the operation commands in advance shown in Fig. 6 has an actuator that operates according to the joint angle command value y indicating the joint angle. It is assumed that the multi-fingered robot node 1 having k joints (k is a positive integer) is controlled.
- the data glove 4 constitutes a remote command device which is mounted on the operator's hand and outputs n (n is a positive integer) output signals xn according to the movement of the operator's hand.
- the control device for a multi-finger bot hand includes a correspondence determining means 6 and a control signal generating means 7.
- the correspondence determining means 6 determines in advance the correspondence between each joint angle command value y and n output signals xn for each of the k joint angle command values y for k joints. It is configured.
- the correspondence relation determining means 6 includes a time-series joint angle pattern storage means 61, a sampling means 62, a partial regression coefficient deriving means 63, and a multiple regression equation determining means 64.
- Sampling means 63 inputs a joint angle command value to k joints (actually, actuator 2) in accordance with the time-series joint angle pattern stored in predetermined time-series joint angle pattern storage means 61.
- the n output signals xn are output m times (m Is a positive integer greater than n).
- the sampling means 62 gives a command to the time-series joint angle pattern storage means 61 before starting sampling, and the time-series joint angle pattern storage means 61 outputs a predetermined joint angle command value to the driving device 3.
- the time-series joint angle pattern is preferably configured to give an arbitrary function command that moves evenly in the N-dimensional space to the joint (actuator) of the multi-fingered robot hand.
- the driving means 3 receives this and causes the multi-fingered robot hand 1 to operate.
- the partial regression coefficient deriving means 63 includes a time-series joint angle pattern stored in the time-series joint angle pattern storage means 61 and n sets of m sets obtained by sampling m times output from the sampling means 62. A multiple regression analysis is performed based on the output signal xn, and one for each of the k joints! / Derive partial regression coefficients.
- the control signal generation means 7 is obtained by the multiple regression equation determination means 64 of the correspondence relation determination means 6.
- the k joint angle command values y are obtained based on the corresponding relationship and the n output signals xn, and a control signal (operation command) indicating the k joint angle command values y is k It is configured to output to a driving device 3 for driving the individual joints.
- the control signal generating means 7 uses the k multiple regression equations determined by the multiple regression equation determination means 64 as a correspondence, and obtains k joint angle command values y by inputting n output signals xn. These k joint angle command values y are operation commands.
- the operation command is output from the control command generation means 7 at a time interval corresponding to the processing speed in accordance with the movement of the data glove 4.
- HANDi ao [n] + ⁇ ai [n] ⁇ GLOVEi [n]
- HAND is a multi-fingered robot hand joint value
- a is a partial regression coefficient
- GLOVE is a multi-fingered robot hand joint value
- [n] is the data globe output with n degrees of freedom, and i is the number of joints of the multi-fingered robot hand.
- the row is larger than the column, and an over-determination is made. Therefore, it is preferable to find the partial regression coefficient as the optimal solution using singular value decomposition.
- FIG. 7 is a flowchart showing an algorithm of software used when the main part of the correspondence determining means 6 of FIG. 6 is realized by using a computer.
- step ST1 a data pair of the data globe 4 and the time-series joint angle pattern is obtained. That is, the multi-fingered robot bot 1 is driven by giving a function that evenly moves the movable range of each joint of the multi-fingered bot nose 1 and the und 1 and the operator's finger performs the same operation in accordance with the multi-fingered robot hand 1. . If the joint angle or joint angle command value ⁇ y of the multi-fingered robot hand 1 is set to y and the output of n data gloves is set to Xn, time series data up to time 11 m can be obtained as shown in the following formula.
- n includes noise such as system noise and quantization error.
- step ST3 estimation of the solution x— by the least squares method is started.
- a partial regression coefficient is obtained.
- the equation used for the least squares method is the following equation.
- step ST7 values of c and ⁇ are determined.
- ⁇ ⁇ ⁇ is calculated by the following matrix
- step ST8 the process proceeds to step ST8, in which [P] t [A] [P] is calculated and set as a new value [A].
- step ST9 the eigenvector V is obtained from the product of the matrix [P] by the following equation.
- V [p] t ( m) [ ⁇ ] "' ⁇ ⁇ [ ⁇ ] ⁇ (2) [ ⁇ ] ⁇ )
- [P] t is the matrix [P] used for the N-th conversion.
- step ST10 the process proceeds to step ST10, and ends if any off-diagonal element of [A] has a convergence condition ⁇ 0.0001. Otherwise, the process returns to step ST5.
- the diagonal elements of the matrix [B] are eigenvalues.
- step ST12 the eigenvalues ( ⁇ , ⁇ ,%) Are changed to ( ⁇ ( ⁇ + ⁇ ), 1 / ( ⁇ + ⁇ ),
- [A] -1 is expressed as follows.
- step ST13 the estimation of the solution X— by the least squares method ends.
- multiple regression analysis is performed using the least squares method and singular value decomposition to derive the partial regression coefficient.
- the above steps ST4 to ST10 are the steps in which the singular value decomposition is performed.
- the pre-collection operation command storage device 8 stores a plurality of data gloves 4 when the hand (moving object) fitted with the data gloves 4 performs a predetermined operation.
- the operation command generated by the operation command generator (6, 7) based on the output of the sensor 5 is stored as a pre-collection operation command.
- Figure 8 shows the software algorithm used to implement this storage operation using a computer.
- the same operation as when the data glove 4 is attached to the hand by the powerful camera 9 such as a monocular camera or a binocular camera to acquire a pre-collection operation command ( (Predetermined operation) is performed by human hand H.
- the image data of the hand (moving object) H at this time is acquired in time series.
- the acquisition of the image data may be performed simultaneously with the acquisition of the pre-collection operation command, or may be performed later.
- image data a state in which a hand “H” has been formed with a hand H is taken, and based on this image data, image data PD1 of a simulated hand is created by computer graphic technology.
- image data in a state of forming a “par” with the hand H is taken, and based on the image data, image data PDn of a simulated hand is created by computer graphic technology. Until it changes from “goo” to “par”
- the image of the process during the process may be actually taken, but the image during the process may be created by combigraphic technology as shown in FIG.
- the image data created in this way is stored in the image data storage device 10.
- each image data included in the plurality of image data obtained in the second step is associated with the pre-collection operation command stored in the It is stored in the corresponding operation command storage means 11.
- the image-related operation command storage means 11 stores the uncreated image data between the previous image data PD1 and the subsequent image data PDn acquired in a time series that includes only the actually photographed image.
- the pre-collection operation command corresponding to the created image data is estimated based on the pre-collection operation command corresponding to the previous image data PD1 and the pre-collection operation command corresponding to the subsequent image data PDn.
- it can be created and stored in association with uncreated image data. In this way, more image data and corresponding data of the pre-collection operation command can be obtained with less image data, and the basic data (data in which the image data is associated with the pre-collection operation command) can be obtained. Collecting becomes much easier.
- the first to third steps are performed as a preliminary operation for driving the robot node. If we try to realize joint angle control with a precision of 5 degrees for each degree of freedom by driving the robot hand, the force that requires about 100 million (20 7) images is required.
- the use of computer graphics technology simplifies the collection of basic data.
- "Poser 5 (Curious Labs Incorporated)" can be used as the CG editing software that can be used in this case. If this software is used, it is possible to interpolate and generate an ideal finger CG image similar to a captured two-dimensional finger image from the joint data obtained from the data globe output.
- the pre-collection operation command corresponding to the interpolated image data can be generated by linear interpolation of the joint angle.
- the file size of finger CG per image is about 5kB.
- the fourth step and the fifth step are steps when actually driving the robot hand.
- a desired operation is performed by hand to operate the robot node.
- the image data of the hand at that time is acquired as time-series robot operation image data using the camera 12.
- This robot operation image data (motion image data Are stored in the image data storage device 13 for robot operation.
- image data corresponding to the motion image data included in the robot motion image data is stored in the image-corresponding motion command storage means 11 and specified by a plurality of image data power time series. This specifying operation is performed by the image data specifying and operation command generator 14.
- the correspondence between the motion image data included in the robot motion image data and the plurality of image data stored in the image-corresponding motion command storage means is determined based on the similarity between the two.
- the method of obtaining the similarity is arbitrary.
- a pre-collection operation command corresponding to the specified image data is given to the driving device 3 as an operation command, and the output of the driving device 3 drives the multi-fingered robot hand.
- the algorithm of the software when the main parts of the fourth and fifth steps are implemented using a computer is as shown in FIG.
- a feature extraction technique using a higher-order local autocorrelation function is used for IJ. it can.
- the higher-order local autocorrelation function is described in detail in a paper entitled “Study on the Application of Statistical Methods for Flexible Information Processing” on pages 126 to 140 of the Research Institute of Electro-Informatics, published in 1993, j957. .
- the higher-order local autocorrelation function was used for feature extraction of the two-dimensional finger image. If the target image in the screen is f (r), the Nth autocorrelation function is defined as follows with respect to the displacement direction (al, a2, —an).
- the order N of the higher-order autocorrelation coefficient was set to 2, and the displacement direction was limited to a local 3 ⁇ 3 pixel area around the reference point r.
- the number of features M is 25 as shown in Fig.11.
- the black square in FIG. 11 corresponds to the position of the corresponding pixel in the local pattern.
- the calculation of each feature is obtained by adding the product of the values of the corresponding pixels of the local pattern to all the pixels.
- Figures 12 (A) to 12 (C) show the different elements of individual differences that appear in human hands.
- Figure 12 (A) shows a state in which the hands are held in a goo shape
- FIG. 12 (B) shows the situation when the hand is shaped like a tyre.
- Figure 12 (C) shows the hand opened in the shape of a par with the difference in the reference angle of the fingers, and in this state, the degree of opening between the thumb and index finger and the curvature of the thumb.
- Figure 13 shows the difference in image data when the curvature of the base position and the degree of overhang of the thumb base are changed by changing the parameters of the computer graphic editing software. By simply changing these two different elements, it is possible to obtain image data with very different individual differences.
- a pre-collection operation command corresponding to the uncreated image data to be created is presumed based on the pre-collection operation command corresponding to the previous image data and the pre-collection operation command corresponding to the subsequent image data, and this is prepared.
- the image data can be stored in association with the uncreated image data.
- the plurality of image data created in the second step may include a plurality of resolution-changed image data created by changing the resolution of the plurality of image data.
- the resolution of image data can be easily changed if there is basic image data.
- Fig. 14 shows three types of image data with different resolutions. The image on the left is the original image, the image at the center is the image with 2 pixels vertically and 2 pixels horizontally and 1 pixel reduced in resolution, and the image on the right is 4 pixels vertically and 4 pixels horizontally. This is an image whose resolution is further reduced as one pixel. For example, if there are 15,000 original images, 45,000 types of image data can be obtained by changing the resolution in two steps.
- the use of a feature extraction technique using a higher-order local autocorrelation function to specify image data was described.
- the feature amount is extracted for the entire image.
- the center of gravity of the hand in the image data is first estimated.
- computer graphic image data is converted into a black and white image. Then, the center of each of the black-and-white images in the X-axis direction and the Y-axis direction is obtained using the following equation.
- xg and yg are the X and Y coordinates of the center of gravity
- xi and yi are the X and Y coordinates of the white pixel
- k is the number of white pixels.
- FIGS. 15 (A) and (B) show a state where one screen composed of 320 ⁇ 240 pixels is divided into 4 ⁇ 4 screens and a state where it is divided into 8 ⁇ 8 screens. If the number of pixels is the above, it is also possible to divide the screen into 16 ⁇ 16 screens.
- the feature amount is calculated for each of the divided images thus divided. That is, a higher-order local autocorrelation pattern is calculated for each screen division.
- the higher-order local autocorrelation pattern for each screen division may be calculated according to the method of obtaining the higher-order local autocorrelation pattern described above.
- the following equation is an equation used for calculating a higher-order local autocorrelation pattern.
- N is 2.
- (Al, a2, -'an) is the displacement direction of the Nth order autocorrelation function
- f (r) is the luminance value of the pixel position r of the target image in the screen. If the local pattern shown in FIG. 12 is used, 25 patterns can be used.
- the following equation is an equation for calculating the principal component score used in the principal component analysis.
- Zkp is the principal component score of data p in the k-th principal component
- X is the n-th feature of the m-th screen at the first resolution of data p (1 is the lowercase letter of the alphabet L) pimn
- a is the factor load of the n-th feature of the m-th screen with the first resolution of the k-th principal component, and klmn
- div is the number of resolutions (for example, 3), and pnum is the number of screen divisions (for example, 8 x 8).
- the principal component score of P can be obtained. Note that the factor loading a
- the klmn calculation method includes a principal factor method, a least squares method, and a maximum likelihood method. To find the factor loading by the main factor method, do the following.
- the contribution ratio of each principal component is obtained to reduce the feature amount.
- the contribution ratio is a coefficient representing the power of how much each principal component explains the original information, and can be expressed as follows. [Number 13]
- C is the contribution ratio of the k-th principal component
- b is the correlation coefficient between the principal component score Zkp and x, defined as follows:
- A is a factor load of the n-th feature amount of the m-th screen of the first resolution of the k-th principal component
- ⁇ is the k-th largest eigenvalue of the correlation matrix.
- the contribution ratio C is
- the component is a coefficient representing the power that explains how much of the original information, and the following relational expression holds.
- the number of principal components used in feature reduction is determined by the following equation with a cumulative contribution ratio of about 95%.
- the data to be searched are the data with the principal component score Zkp that is the same as or the closest to the unknown image in each main component, and the number of before and after data according to the contribution ratio of each main component.
- the number of candidates for each principal component is as follows.
- Cc is the total sum of the number of estimated candidates, which is preliminarily determined.
- fractional expression including ⁇ ⁇ is the contribution ratio of the ⁇ component in the tenth principal component.
- the above equation determines that the number of preceding and succeeding data corresponding to the contribution ratio of each principal component is to be searched. For example, referring to FIG. 16, assuming that the principal component score Zkp of the first principal component of the motion image data to be collated is 13 points, sorting is performed in the order of the principal component score of the first principal component.
- the image data of the above-mentioned estimated number of candidates is extracted as the image data for comparison, mainly including the image data whose principal component score is close to 13 points among the base image data.
- FIG. 17 shows a procedure for creating a database of image data including the above-described improvements (corresponding to the first to third steps of the embodiment).
- FIG. 18 shows a specific flow of the collation procedure.
- a processing speed higher than the video rate can be achieved, and thus a high-speed camera can be used for imaging.
- the imaging power also removes the background to remove unnecessary information. For this purpose, a process of subtracting (background image) from imaging of (hand image + background image) is performed.
- the feature amount of the captured image (moving image data) is calculated.
- the feature amount may be calculated as it is when the image was captured, but in the example of FIG. 18, three (multiple) types of motion image data having different resolutions are obtained by changing the resolution of the captured motion image data. In other words, simply create three times as much data as the same motion image data.
- the center of gravity of the hand image is estimated and the screen is divided in the same manner as when the database was created. Also in this case, the screen is divided into 16, 64 or 246.
- the number of divisions is arbitrary. However, it has been confirmed that increasing the number of divisions does not increase the accuracy, so a division number of about 16 divisions or 64 divisions is preferred.
- a higher-order local autocorrelation pattern is calculated for each screen division.
- the calculation of the higher-order local autocorrelation pattern at this time is the same as the calculation method of the higher-order local autocorrelation pattern used when creating the above-mentioned database.
- a principal component score is calculated by principal component analysis based on the calculated higher-order local autocorrelation pattern.
- the calculation of the principal component score also employs the same calculation method as the calculation method of the principal component score when the database is created.
- a plurality of collation image data which are candidates for collation are selected based on the principal component scores obtained by calculation.
- the number of candidates for each principal component force is determined in advance by the above equation [Equation 17]. Therefore, the image data having the closest principal component score Zkp in each principal component and the image data before and after this image data are selected for the predetermined number of candidates according to the contribution ratio of each principal component, and are selected for comparison.
- Image data. 10 sorts based on the 10 principal components stored in the database
- fi (x) is the principal component score of the ith principal component for which the feature power has also been calculated
- X is a feature value of the candidate r based on a higher-order local autocorrelation function
- X is a feature quantity by a higher-order local autocorrelation function at time t
- the data p that minimizes Er is the search image, and the joint angle data of the data p is the estimated angle.
- the search at time t ends. If a significantly different joint angle is selected from time t1, the next candidate with the next smallest Euclidean distance Er is selected, and whether or not it is within the allowable range is calculated by the following equation.
- ang is the i-th joint angle data at time t.
- the image data is specified as corresponding to the operation image data.
- FIG. 19 is a flowchart showing a procedure for creating a database in advance
- FIG. 21 is a flowchart showing details of the flowchart of FIG.
- the video camera 109 captures an image of a hand wearing gloves.
- a database is constructed using hand image data captured by the high-speed video camera 109 and angle data obtained from the data glove force.
- image data and angle data are acquired as a set.
- a monochrome high-speed camera ME310, manufactured by MEGAPLUS, ES310ZT
- the resolution is 320 * 240 pixels
- the finger is sufficiently large on the screen.
- a data glove called "CyberGlo Ve " (trademark) manufactured by Virtual Technologies was used.
- the gloves were all white and plain.
- the gloves are fitted over the data gloves to prevent data glove-specific image features (sensor shape and glove shape) from appearing.
- the color is different from that of a human hand, and the image is also different.
- the image data can be handled as a general hand image by the feature amount as described later.
- a primary database is created by combining the finger angle data acquired with the data glove force together with a plurality of continuous image data. To obtain the necessary and sufficient information, move the hand continuously to acquire and save the required image data and angle data at the same time.
- image processing for converting the image data of the primary database into a feature amount necessary for estimating the actually shot hand image is performed. For this image processing, Then, change the resolution, extract edges, and extract contours (feature extraction).
- the resolution of the image from which the background has been removed is changed.
- the image acquired this time is an image with a pixel value of 320 X 240. It takes a lot of time to perform estimation processing by applying various image processing using this as it is. Further, when a human estimates, it is possible to estimate even if the size of the image is smaller than that used in the present embodiment. Therefore, in the present embodiment, the resolution is changed from a pixel value of 320 ⁇ 240 to a size of 64 ⁇ 64.
- the pixel value after changing the resolution is obtained by the following equation.
- gr (i, j) is the pixel value of the i-th row and the j-th column after the resolution is changed.
- go (i, j) is the pixel value at the i-th row and the j-th column before the resolution is changed.
- the vertical is also calculated at 320. This is because the pixel value after the change is 64 X 64, so that the aspect ratio is adjusted.
- the calculation is performed up to (i + 1) ⁇ 320Z64 ⁇ 1, where r is the number of k X I.
- Edge extraction is performed using the resolution-changed image obtained by the above operation. In this work, features necessary for estimation are extracted. Specifically, the resolution-changed image is filtered by a Sobel filter. Thereafter, edges are extracted by binarizing only the outermost periphery of this image.
- fr (i, j) can be represented by the following equation.
- fi (i, j) represents the i-th horizontal and half-th pixel before processing.
- f s is the final value after processing.
- the binarization of the outermost periphery of the image here means that the image is binarized at a relatively low threshold so that the outline is not interrupted by the outline extraction.
- contour extraction is performed by scanning the binary image obtained by edge extraction.
- the reason why the image obtained by the edge extraction in the previous stage is not used as a contour is that the change in ambient light can be increased. There are differences in the ambient light, human skin color, etc. between the actual estimation and the database creation. Furthermore, since white gloves are used when creating the database, even if edge extraction is performed, the edge value and width will differ from the actual image edges.
- FIG. 22 shows an image after resolution change obtained by performing edge extraction after edge extraction.
- the image was scanned from above and the first 127 points were set as the starting points.
- the starting point also searches for neighboring pixels counterclockwise. If there is a point of 127, move to that point and rewrite the value to 255. After moving, the search is continued by dividing the search start point into eight patterns with reference to the previous position. When it returns to the starting point again, it is determined that the contour has been extracted, and the search is terminated. Finally, since the remaining points are not contours, all 127 points are set to 0.
- the local autocorrelation feature is a feature that is effective for image recognition and measurement, and is obtained by calculating a reference point and its vicinity with respect to a higher-order correlation function defined by the following equation. .
- the image screen is divided into eight vertically and eight horizontally.
- the dimension was reduced to 25 feature amounts by a higher-order local autocorrelation function using all the pixels as reference points. That is, considering the whole image, the feature amount of the whole image can be converted into the feature amount of the 25 X screen division number.
- the feature data which is a combination of the image data obtained by performing the feature amount of one image data and the angle data (corresponding to the pre-collection operation command) from the data glove, is stored in the image corresponding operation command storage means (database ).
- the image corresponding operation command storage means database .
- an image actually shot is specified by using this database.
- FIG. 25 is an algorithm showing a procedure at the time of estimation
- FIG. 26 is an algorithm in more detail.
- an image (moving image data) of a hand moving with a high-speed camera is acquired as in the case of creating a database.
- the obtained motion image data is subjected to the same image processing as that at the time of database creation to extract a feature amount.
- the feature amount of the motion image data and the feature amounts of a plurality of image data stored in the database are compared by a process described later, and the angle data that is paired with the image data closest to the motion image data is calculated.
- Output as an operation command corresponding to the image data.
- X is a feature amount i of the candidate r based on the higher-order local autocorrelation function
- X is a feature amount i based on the higher-order local autocorrelation function at the time t nti.
- the angle of the data r that minimizes Er was set as the estimated angle of the motion image data.
- the angle is limited by comparing it with the previous estimated angle (time t1). Specifically, when the evaluation function Ap shown by the following equation exceeds a specified value for a short distance, the estimation candidate is excluded. If the evaluation function Ap is within the specified value, the estimated angle is adopted.
- ang is the i-th angle data at time t. Also, i (t) obtained from the data glove
- the angle information is 24.
- the current estimated angle is obtained by the above operation.
- FIG. 27 is a conceptual diagram for explaining clustering.
- one row indicates one set composed of a plurality of pieces of image data.
- the darkly colored rows are the data to be searched first, and the lightly colored rows are the data to be searched in detail. If the data included in the database is created using the algorithm for creating the database described above, hand-shaped data will come together in the image data sequence. By utilizing this, the target image data for distance calculation is reduced. Specifically, a plurality of image data is considered as one set. Then, one image data to be used at the time of the first distance calculation is selected from each of a plurality of sets, and the distance calculation is performed with the selected image data.
- the calculation cost of the distance calculation is reduced by 1 for each set, and the cost is greatly reduced for all the sets.
- the accuracy is lower than when the estimation result is compared with all the image data in the original database. Therefore, in the distance calculation in the first step, for those having a short distance, the distance between all the image data in the set including each image data and the motion image data is calculated. In this way, since data near angles are gathered near each data, a more accurate estimation result can be obtained.
- ten image data were considered as one set, and the first-stage distance calculation was performed for one motion image data with one image data in multiple sets.
- the same operation as that of the moving object or the simulated object without using the sensor is performed by the defect setting only by obtaining an image of the moving object or the simulated object corresponding to the robot. It is possible to make the robot imitate the motion by eliminating the nature. Therefore, according to the present invention, it is possible to easily drive a robot using image data as input data.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Robotics (AREA)
- Mechanical Engineering (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Automation & Control Theory (AREA)
- Manipulator (AREA)
- Image Analysis (AREA)
Description
Claims
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US10/595,810 US7848850B2 (en) | 2003-11-13 | 2004-11-15 | Method for driving robot |
DE112004002219.3T DE112004002219B4 (de) | 2003-11-13 | 2004-11-15 | Verfahren zum Antreiben eines Roboters |
GB0611135A GB2424723B (en) | 2003-11-13 | 2004-11-15 | Method for driving robot |
JP2005515481A JP4878842B2 (ja) | 2003-11-13 | 2004-11-15 | ロボットの駆動方法 |
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2003-384402 | 2003-11-13 | ||
JP2003384402 | 2003-11-13 | ||
JP2004-173268 | 2004-06-10 | ||
JP2004173268 | 2004-06-10 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2005046942A1 true WO2005046942A1 (ja) | 2005-05-26 |
Family
ID=34593956
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2004/016968 WO2005046942A1 (ja) | 2003-11-13 | 2004-11-15 | ロボットの駆動方法 |
Country Status (5)
Country | Link |
---|---|
US (1) | US7848850B2 (ja) |
JP (2) | JP4878842B2 (ja) |
DE (1) | DE112004002219B4 (ja) |
GB (1) | GB2424723B (ja) |
WO (1) | WO2005046942A1 (ja) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPS6261095A (ja) * | 1985-09-11 | 1987-03-17 | カシオ計算機株式会社 | 波形エディット装置 |
JP2008080431A (ja) * | 2006-09-27 | 2008-04-10 | Advanced Telecommunication Research Institute International | ロボットシステム |
JP2008140267A (ja) * | 2006-12-04 | 2008-06-19 | National Institute Of Advanced Industrial & Technology | 動作認識装置および動作認識処理方法 |
WO2009147904A1 (ja) | 2008-06-04 | 2009-12-10 | 国立大学法人筑波大学 | 手指形状推定装置、手指形状の推定方法及びプログラム |
JP2010089248A (ja) * | 2008-10-09 | 2010-04-22 | National Chiao Tung Univ | ハンドパペットマニピュレーションシステム |
WO2011039429A1 (fr) | 2009-09-29 | 2011-04-07 | Artistic Robot Event | Systeme et procede de restitution synchrone par un robot de mouvements d'une main ou d'un organe d'extremite |
WO2011065034A1 (ja) * | 2009-11-24 | 2011-06-03 | 株式会社豊田自動織機 | ロボットの動作を制御する方法およびロボットシステム |
WO2011065035A1 (ja) * | 2009-11-24 | 2011-06-03 | 株式会社豊田自動織機 | ロボットの教示データを作成する方法およびロボット教示システム |
JP2011121152A (ja) * | 2009-12-11 | 2011-06-23 | Toyota Motor Corp | 動作補助装置 |
WO2012117687A1 (ja) * | 2011-03-02 | 2012-09-07 | パナソニック株式会社 | 姿勢推定装置、姿勢推定システム、および姿勢推定方法 |
JP2014140941A (ja) * | 2013-01-25 | 2014-08-07 | Seiko Epson Corp | ロボット制御システム、ロボット、ロボット制御方法及びプログラム |
JP2014140942A (ja) * | 2013-01-25 | 2014-08-07 | Seiko Epson Corp | ロボット制御システム、ロボット、ロボット制御方法及びプログラム |
CN112847334A (zh) * | 2020-12-16 | 2021-05-28 | 北京无线电测量研究所 | 一种基于视觉伺服的机械臂目标跟踪方法 |
Families Citing this family (41)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7668797B2 (en) * | 2006-04-07 | 2010-02-23 | Gary Kuvich | Active semiotic system for image and video understanding by robots and unmanned vehicles, methods and apparatus |
JP5148619B2 (ja) * | 2006-10-18 | 2013-02-20 | ユタカ・ジェイ・カナヤマ | 移動ロボットによる地図作成方法 |
US9060385B1 (en) * | 2007-04-20 | 2015-06-16 | Lloyd Douglas Manning | Universal bluetooth/wireless glove |
JP5109098B2 (ja) * | 2007-06-14 | 2012-12-26 | 本田技研工業株式会社 | 運動制御システム、運動制御方法および運動制御プログラム |
US8213706B2 (en) * | 2008-04-22 | 2012-07-03 | Honeywell International Inc. | Method and system for real-time visual odometry |
KR101494344B1 (ko) * | 2008-04-25 | 2015-02-17 | 삼성전자주식회사 | 휴머노이드 로봇의 움직임 제어 시스템 및 그 방법 |
KR100995933B1 (ko) * | 2008-09-01 | 2010-11-22 | 한국과학기술연구원 | 진화 알고리즘과 모방학습에 기초한 로봇의 동작 제어 방법 |
TWI371303B (en) * | 2009-02-04 | 2012-09-01 | Inventec Appliances Corp | Electronic pet cotrol system and control method thereof |
US9052710B1 (en) * | 2009-03-20 | 2015-06-09 | Exelis Inc. | Manipulation control based upon mimic of human gestures |
US8260460B2 (en) * | 2009-09-22 | 2012-09-04 | GM Global Technology Operations LLC | Interactive robot control system and method of use |
WO2011127410A2 (en) * | 2010-04-09 | 2011-10-13 | Deka Products Limited Partnership | System and apparatus for robotic device and methods of using thereof |
JP5306313B2 (ja) * | 2010-12-20 | 2013-10-02 | 株式会社東芝 | ロボット制御装置 |
KR101896473B1 (ko) * | 2012-01-04 | 2018-10-24 | 삼성전자주식회사 | 로봇 핸드의 제어 방법 |
US9070004B2 (en) * | 2012-05-03 | 2015-06-30 | General Electric Company | Automatic segmentation and characterization of cellular motion |
JP5426719B2 (ja) * | 2012-05-18 | 2014-02-26 | ファナック株式会社 | ロボットシステムの動作シミュレーション装置 |
US9393695B2 (en) | 2013-02-27 | 2016-07-19 | Rockwell Automation Technologies, Inc. | Recognition-based industrial automation control with person and object discrimination |
US9798302B2 (en) | 2013-02-27 | 2017-10-24 | Rockwell Automation Technologies, Inc. | Recognition-based industrial automation control with redundant system input support |
US9498885B2 (en) | 2013-02-27 | 2016-11-22 | Rockwell Automation Technologies, Inc. | Recognition-based industrial automation control with confidence-based decision support |
US9804576B2 (en) | 2013-02-27 | 2017-10-31 | Rockwell Automation Technologies, Inc. | Recognition-based industrial automation control with position and derivative decision reference |
JP6271903B2 (ja) * | 2013-08-06 | 2018-01-31 | キヤノン株式会社 | サーボ装置、及びサーボ装置の制御方法 |
KR102000264B1 (ko) * | 2013-10-01 | 2019-07-15 | 한국전자통신연구원 | 교시 데이터 입력 장치와 이를 이용한 로봇의 교시 명령어 생성 장치 및 방법 |
US9084411B1 (en) * | 2014-04-10 | 2015-07-21 | Animal Biotech Llc | Livestock identification system and method |
JP6003942B2 (ja) * | 2014-04-24 | 2016-10-05 | トヨタ自動車株式会社 | 動作制限装置及び動作制限方法 |
US10970428B2 (en) * | 2015-04-01 | 2021-04-06 | Research & Business Foundation Sungkyunkwan University | Method of providing performance indicators of robot hand and method of optimizing structure of robot hand |
US9694494B1 (en) | 2015-12-11 | 2017-07-04 | Amazon Technologies, Inc. | Feature identification and extrapolation for robotic item grasping |
US10456910B2 (en) * | 2016-01-14 | 2019-10-29 | Purdue Research Foundation | Educational systems comprising programmable controllers and methods of teaching therewith |
DE112016006116T5 (de) * | 2016-01-29 | 2018-09-13 | Mitsubishi Electric Corporation | Roboterlehrvorrichtung und Verfahren zum Erzeugen eines Robotersteuerprogramms |
US20170249561A1 (en) * | 2016-02-29 | 2017-08-31 | GM Global Technology Operations LLC | Robot learning via human-demonstration of tasks with force and position objectives |
US11036230B1 (en) * | 2016-03-03 | 2021-06-15 | AI Incorporated | Method for developing navigation plan in a robotic floor-cleaning device |
US9827670B1 (en) * | 2016-05-24 | 2017-11-28 | X Development Llc | Coaxial finger face and base encoding |
TW201805598A (zh) * | 2016-08-04 | 2018-02-16 | 鴻海精密工業股份有限公司 | 自主移動設備及建立導航路徑的方法 |
JP6514156B2 (ja) * | 2016-08-17 | 2019-05-15 | ファナック株式会社 | ロボット制御装置 |
US10319109B2 (en) * | 2017-03-31 | 2019-06-11 | Honda Motor Co., Ltd. | Interaction with physical objects as proxy objects representing virtual objects |
CN107199566B (zh) * | 2017-06-02 | 2019-09-10 | 东南大学 | 一种基于虚拟手臂的面向空间站机器人的遥操作系统 |
DE112018002565B4 (de) * | 2017-08-10 | 2021-07-01 | Robert Bosch Gmbh | System und Verfahren zum direkten Anlernen eines Roboters |
WO2019064752A1 (ja) * | 2017-09-28 | 2019-04-04 | 日本電産株式会社 | ロボット教示システム、ロボット教示方法、制御装置、及びコンピュータプログラム |
WO2019064751A1 (ja) * | 2017-09-28 | 2019-04-04 | 日本電産株式会社 | ロボット教示システム、ロボット教示方法、制御装置、及びコンピュータプログラム |
DE102018124671B4 (de) * | 2018-10-06 | 2020-11-26 | Bystronic Laser Ag | Verfahren und Vorrichtung zur Erstellung eines Robotersteuerprogramms |
DE102020124285B4 (de) | 2019-09-20 | 2022-06-09 | Nvidia Corporation | Visionsbasierte Teleoperation eines beweglichen Robotersystems |
US20220080581A1 (en) * | 2020-09-11 | 2022-03-17 | Fanuc Corporation | Dual arm robot teaching from dual hand human demonstration |
US20230109729A1 (en) * | 2021-10-07 | 2023-04-13 | Nec Laboratories America, Inc. | Co-disentagled series/text multi-modal representation learning for controllable generation |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2000329511A (ja) * | 1999-05-19 | 2000-11-30 | Sony Corp | データグローブ及びこれを用いた形状認識方法 |
JP2001322079A (ja) * | 2000-05-15 | 2001-11-20 | Sony Corp | 脚式移動ロボット及びその動作教示方法 |
JP2003080484A (ja) * | 2001-09-07 | 2003-03-18 | Tomy Co Ltd | 動作反応玩具 |
Family Cites Families (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPS61122711A (ja) * | 1984-11-20 | 1986-06-10 | Fujitsu Ltd | ロボツトのテイ−チング装置 |
JPH01322079A (ja) * | 1988-06-22 | 1989-12-27 | Shibuya Kinzoku Sangyo Kk | 修正装置付ピボットヒンジ |
JPH0380484A (ja) * | 1989-08-22 | 1991-04-05 | Hitachi Maxell Ltd | テープカートリッジ |
US5887069A (en) * | 1992-03-10 | 1999-03-23 | Hitachi, Ltd. | Sign recognition apparatus and method and sign translation system using same |
US6072494A (en) * | 1997-10-15 | 2000-06-06 | Electric Planet, Inc. | Method and apparatus for real-time gesture recognition |
US6236037B1 (en) | 1998-02-20 | 2001-05-22 | Massachusetts Institute Of Technology | Finger touch sensors and virtual switch panels |
US6421453B1 (en) * | 1998-05-15 | 2002-07-16 | International Business Machines Corporation | Apparatus and methods for user recognition employing behavioral passwords |
US6681031B2 (en) * | 1998-08-10 | 2004-01-20 | Cybernet Systems Corporation | Gesture-controlled interfaces for self-service machines and other applications |
JP4340367B2 (ja) * | 1999-01-27 | 2009-10-07 | 株式会社リコー | 画像分類装置およびその装置としてコンピュータを機能させるためのプログラムを記録したコンピュータ読み取り可能な記録媒体 |
JP2001092589A (ja) * | 1999-09-22 | 2001-04-06 | Suzuki Motor Corp | モデリング装置及びモデリングデータ生成用プログラムを記憶した記憶媒体 |
JP3457617B2 (ja) * | 2000-03-23 | 2003-10-20 | 株式会社東芝 | 画像検索システムおよび画像検索方法 |
JP2001344567A (ja) * | 2000-06-02 | 2001-12-14 | Ricoh Co Ltd | 文字認識装置、その方法およびその方法を実施するためのプログラムを記録した記録媒体 |
JP2002163655A (ja) * | 2000-11-24 | 2002-06-07 | Omron Corp | 個人認証装置 |
JP2002239960A (ja) * | 2001-02-21 | 2002-08-28 | Sony Corp | ロボット装置の動作制御方法、プログラム、記録媒体及びロボット装置 |
US6804396B2 (en) | 2001-03-28 | 2004-10-12 | Honda Giken Kogyo Kabushiki Kaisha | Gesture recognition system |
US20020152077A1 (en) * | 2001-04-12 | 2002-10-17 | Patterson Randall R. | Sign language translator |
JP4028192B2 (ja) * | 2001-08-02 | 2007-12-26 | 日本電信電話株式会社 | 動作入力方法、装置、プログラム、及び記録媒体 |
JP3848123B2 (ja) | 2001-10-15 | 2006-11-22 | 独立行政法人科学技術振興機構 | 人型ロボットハンド |
JP2003288594A (ja) * | 2002-03-28 | 2003-10-10 | Minolta Co Ltd | 画像処理装置および方法 |
US7665041B2 (en) * | 2003-03-25 | 2010-02-16 | Microsoft Corporation | Architecture for controlling a computer using hand gestures |
US7565295B1 (en) * | 2003-08-28 | 2009-07-21 | The George Washington University | Method and apparatus for translating hand gestures |
-
2004
- 2004-11-15 GB GB0611135A patent/GB2424723B/en not_active Expired - Fee Related
- 2004-11-15 WO PCT/JP2004/016968 patent/WO2005046942A1/ja active Application Filing
- 2004-11-15 JP JP2005515481A patent/JP4878842B2/ja not_active Expired - Fee Related
- 2004-11-15 DE DE112004002219.3T patent/DE112004002219B4/de not_active Expired - Fee Related
- 2004-11-15 US US10/595,810 patent/US7848850B2/en not_active Expired - Fee Related
-
2011
- 2011-03-22 JP JP2011062873A patent/JP5209751B2/ja not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2000329511A (ja) * | 1999-05-19 | 2000-11-30 | Sony Corp | データグローブ及びこれを用いた形状認識方法 |
JP2001322079A (ja) * | 2000-05-15 | 2001-11-20 | Sony Corp | 脚式移動ロボット及びその動作教示方法 |
JP2003080484A (ja) * | 2001-09-07 | 2003-03-18 | Tomy Co Ltd | 動作反応玩具 |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPS6261095A (ja) * | 1985-09-11 | 1987-03-17 | カシオ計算機株式会社 | 波形エディット装置 |
JP2008080431A (ja) * | 2006-09-27 | 2008-04-10 | Advanced Telecommunication Research Institute International | ロボットシステム |
JP2008140267A (ja) * | 2006-12-04 | 2008-06-19 | National Institute Of Advanced Industrial & Technology | 動作認識装置および動作認識処理方法 |
WO2009147904A1 (ja) | 2008-06-04 | 2009-12-10 | 国立大学法人筑波大学 | 手指形状推定装置、手指形状の推定方法及びプログラム |
US9002119B2 (en) | 2008-06-04 | 2015-04-07 | University Of Tsukuba, National University Corporation | Device method and program for human hand posture estimation |
US8571707B2 (en) | 2008-10-09 | 2013-10-29 | National Chiao Tung University | Glove puppet manipulating system |
JP2010089248A (ja) * | 2008-10-09 | 2010-04-22 | National Chiao Tung Univ | ハンドパペットマニピュレーションシステム |
WO2011039429A1 (fr) | 2009-09-29 | 2011-04-07 | Artistic Robot Event | Systeme et procede de restitution synchrone par un robot de mouvements d'une main ou d'un organe d'extremite |
WO2011065034A1 (ja) * | 2009-11-24 | 2011-06-03 | 株式会社豊田自動織機 | ロボットの動作を制御する方法およびロボットシステム |
WO2011065035A1 (ja) * | 2009-11-24 | 2011-06-03 | 株式会社豊田自動織機 | ロボットの教示データを作成する方法およびロボット教示システム |
JP2011110621A (ja) * | 2009-11-24 | 2011-06-09 | Toyota Industries Corp | ロボットの教示データを作成する方法およびロボット教示システム |
JP2011110620A (ja) * | 2009-11-24 | 2011-06-09 | Toyota Industries Corp | ロボットの動作を制御する方法およびロボットシステム |
JP2011121152A (ja) * | 2009-12-11 | 2011-06-23 | Toyota Motor Corp | 動作補助装置 |
JP2012181736A (ja) * | 2011-03-02 | 2012-09-20 | Panasonic Corp | 姿勢推定装置、姿勢推定システム、および姿勢推定方法 |
WO2012117687A1 (ja) * | 2011-03-02 | 2012-09-07 | パナソニック株式会社 | 姿勢推定装置、姿勢推定システム、および姿勢推定方法 |
US9480417B2 (en) | 2011-03-02 | 2016-11-01 | Panasonic Corporation | Posture estimation device, posture estimation system, and posture estimation method |
JP2014140941A (ja) * | 2013-01-25 | 2014-08-07 | Seiko Epson Corp | ロボット制御システム、ロボット、ロボット制御方法及びプログラム |
JP2014140942A (ja) * | 2013-01-25 | 2014-08-07 | Seiko Epson Corp | ロボット制御システム、ロボット、ロボット制御方法及びプログラム |
CN112847334A (zh) * | 2020-12-16 | 2021-05-28 | 北京无线电测量研究所 | 一种基于视觉伺服的机械臂目标跟踪方法 |
CN112847334B (zh) * | 2020-12-16 | 2022-09-23 | 北京无线电测量研究所 | 一种基于视觉伺服的机械臂目标跟踪方法 |
Also Published As
Publication number | Publication date |
---|---|
JP5209751B2 (ja) | 2013-06-12 |
JPWO2005046942A1 (ja) | 2007-08-23 |
US7848850B2 (en) | 2010-12-07 |
GB0611135D0 (en) | 2006-07-19 |
GB2424723A (en) | 2006-10-04 |
DE112004002219T5 (de) | 2006-11-23 |
JP2011131376A (ja) | 2011-07-07 |
JP4878842B2 (ja) | 2012-02-15 |
US20070078564A1 (en) | 2007-04-05 |
DE112004002219B4 (de) | 2018-07-12 |
GB2424723B (en) | 2007-09-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP4878842B2 (ja) | ロボットの駆動方法 | |
Newbury et al. | Deep learning approaches to grasp synthesis: A review | |
CN108972494B (zh) | 一种仿人机械手抓取控制系统及其数据处理方法 | |
CN106346485B (zh) | 基于人手运动姿态学习的仿生机械手的非接触式控制方法 | |
Varley et al. | Generating multi-fingered robotic grasps via deep learning | |
Hasan et al. | RETRACTED ARTICLE: Static hand gesture recognition using neural networks | |
Lopes et al. | Visual learning by imitation with motor representations | |
Qu et al. | Human-like coordination motion learning for a redundant dual-arm robot | |
Jia et al. | Manipulating highly deformable materials using a visual feedback dictionary | |
Liu et al. | Hybrid robotic grasping with a soft multimodal gripper and a deep multistage learning scheme | |
Koganti et al. | Bayesian nonparametric learning of cloth models for real-time state estimation | |
Schröder et al. | Real-time hand tracking using synergistic inverse kinematics | |
CN109766782B (zh) | 基于svm的实时肢体动作识别方法 | |
Ottenhaus et al. | Visuo-haptic grasping of unknown objects based on gaussian process implicit surfaces and deep learning | |
Nölker et al. | GREFIT: Visual recognition of hand postures | |
Skoglund et al. | Programming-by-Demonstration of reaching motions—A next-state-planner approach | |
Wang et al. | A generative human-robot motion retargeting approach using a single depth sensor | |
Kumar et al. | Graph matching based hand posture recognition using neuro-biologically inspired features | |
Acosta-Calderon et al. | Robot imitation: Body schema and body percept | |
Gutzeit et al. | Automatic Detection and Recognition of Human Movement Patterns in Manipulation Tasks. | |
Hoshino et al. | Copycat hand—robot hand imitating human motions at high speed and with high accuracy | |
Infantino et al. | A cognitive architecture for robotic hand posture learning | |
Srinivasa et al. | A bio-inspired kinematic controller for obstacle avoidance during reaching tasks with real robots | |
Rodriguez et al. | Learning postural synergies for categorical grasping through shape space registration | |
JP2006294018A (ja) | データベースの高速検索方法及び該高速検索方法を用いたロボットの駆動方法 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AK | Designated states |
Kind code of ref document: A1 Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BW BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EC EE EG ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NA NI NO NZ OM PG PH PL PT RO RU SC SD SE SG SK SL SY TJ TM TN TR TT TZ UA UG US UZ VC VN YU ZA ZM ZW |
|
AL | Designated countries for regional patents |
Kind code of ref document: A1 Designated state(s): BW GH GM KE LS MW MZ NA SD SL SZ TZ UG ZM ZW AM AZ BY KG KZ MD RU TJ TM AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IS IT LU MC NL PL PT RO SE SI SK TR BF BJ CF CG CI CM GA GN GQ GW ML MR NE SN TD TG |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
DPEN | Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed from 20040101) | ||
WWE | Wipo information: entry into national phase |
Ref document number: 2005515481 Country of ref document: JP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 1120040022193 Country of ref document: DE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 0611135.5 Country of ref document: GB Ref document number: 0611135 Country of ref document: GB |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2007078564 Country of ref document: US Ref document number: 10595810 Country of ref document: US |
|
122 | Ep: pct application non-entry in european phase | ||
WWP | Wipo information: published in national office |
Ref document number: 10595810 Country of ref document: US |
|
REG | Reference to national code |
Ref country code: DE Ref legal event code: 8607 |