Fanuc Corporation

Preferred Networks, Inc.

Hidetoshi Kumiya

Kazunori Ban

Katsutoshi Takizawa

Kazutaka Nakayama

Hiroshi Nakagawa

Ryuichi Hara

Daisuke Okanohara

Eiichi Matsumoto

Ryosuke Okuta

Keigo Kawaai

Takumi Oyama

Shun Suyama

Takashi Yamazaki

Atsushi Watanabe

Ichiro Kanno

Shouta Takizawa

Kyouhei Kokubo

Toshiyuki Ando

Atsuo Nagayama

A machine learning device that learns an operation of a robot for picking up, by a hand unit, any of a plurality of objects placed in a random fashion, including a bulk-loaded state, includes a state variable observation unit that observes a state variable representing a state of the robot, including data output from a three-dimensional measuring device that obtains a three-dimensional map for each object, an operation result obtaining unit that obtains a result of a picking operation of the robot for picking up the object by the hand unit, and a learning unit that learns a manipulated variable including command data for commanding the robot to perform the picking operation of the object, in association with the state variable of the robot and the result of the picking operation, upon receiving output from the state variable observation unit and output from the operation result obtaining unit.

Machine learning device, robot system, and machine learning method for learning object picking operation

A machine learning device that learns an operation of a robot for picking up, by a hand unit, any of a plurality of workpieces placed in a random fashion, including a bulk-loaded state, includes a state variable observation unit that observes a state variable representing a state of the robot, including data output from a three-dimensional measuring device that obtains a three-dimensional map for each workpiece, an operation result obtaining unit that obtains a result of a picking operation of the robot for picking up the workpiece by the hand unit, and a learning unit that learns a manipulated variable including command data for commanding the robot to perform the picking operation of the workpiece, in association with the state variable of the robot and the result of the picking operation, upon receiving output from the state variable observation unit and output from the operation result obtaining unit.

Machine learning device, robot system, and machine learning method for learning workpiece picking operation

An article take-out apparatus comprising: a robot having a hand capable of holding an article; a 3D measuring device measuring surface positions of a plurality of articles stored in bulk in a 3D space to acquire a 3D point set composed of a plurality of 3D points; a local maximum point selecting unit selecting from the 3D point set a 3D point, where a coordinate value with respect to a predetermined coordinate axis is maximum; a processing unit determining a hand position and posture including a target position and target posture of the hand enabling an article near the local maximum point to be taken out based on the selected local maximum point; and a robot control unit controlling the robot so as to move the hand to the determined hand position and posture and take out the article from the hand position and posture.

Apparatus and method of taking out bulk stored articles by robot

A method and an apparatus for predicting interference, at practical accuracy and calculation time, between a target section of a robot and a peripheral object installed around the robot, when the target section, such as a tool or a sensor attached to a robot hand, is moved along a movement path thereof due to the motion of the robot. A convex hull, defined by areas occupied by the tool at adjacent time points, is calculated. It is judged whether a common area exists between the convex hull and a polyhedron area. When the common area exists, it is judged that the tool interferes with the container box on the movement path, and the procedure is terminated. When the common area does not exist, it is judged whether j&#x3c;n is true. If j&#x2267;n is true, it is judged that the interference does not occur.

Method and apparatus for predicting interference between target section of robot and peripheral object

A workpiece removing device including a camera for imaging a workpiece loading area including a plurality of workpieces loaded in bulk; a workpiece detection section for detecting a workpiece, based on a camera image taken with the camera; a workpiece selection section for selecting a workpiece adapted to be removed, based on a detection result by the workpiece detection section; a robot for removing the workpiece selected by the workpiece selection section; a loading state determination section for determining whether a loading state of the workpieces in the workpiece loading area has changed due to a operation of the robot; and an area setting section for setting a workpiece detection area where the workpiece detection section detects a workpiece. If the loading state determination section determines that the loading state of the workpieces has changed, the area setting section sets the workpiece detection area in a peripheral area of a changing position of the loading state, i.e., in a portion of the workpiece loading area.

Workpiece removing device and method

An apparatus for picking up objects including a robot for picking up an object, at least one part of the object having a curved shape, having a storing means for storing a gray gradient distribution model of the object, a recognizing means for recognizing a gray image of the object, a gradient extracting means for extracting a gray gradient distribution from the gray image recognized by the recognizing means, an object detecting means for detecting a position or position posture of the object in the gray image in accordance with the gray gradient distribution extracted by the gradient extracting means and the gray gradient distribution model stored by the storing means, a detection information converting means for converting information of the position or position posture detected by the object detecting means into information of position or position posture in a coordinate system regarding the robot; and a robot moving means for moving the robot to the position or position posture converted by the detection information converting means to pick up the object. Thus, the object having a curved shape can be detected and picked up in a reasonably short period of time.

Apparatus for picking up objects

A first masking image of an input image is created by assuming that pixels in an area including the area of a target in all sample images are valid pixels while pixels in the remaining area are invalid pixels. The first masking image is used in a first comparison between the input image and each sample image to select an image that provides a best match as a provisionally selected sample image. In a second comparison, a second masking image corresponding to the provisionally selected sample image is used to select a final sample image. The second masking image is created from an individual sample image by assuming that pixels in an area including the area of the target are valid pixels while pixels in the remaining area invalid pixels.

Image processing apparatus

(1) Orthogonal transformation is applied to sample images based on the images of a normalized orthogonal system, and component groups are determined. A plurality of comparison target images are selected from an input image, and a partial normalized orthogonal system is created from the normalized orthogonal system by decreasing the dimensions. (2) Based on the partial normalized orthogonal system, orthogonal transformation is applied to the comparison target images, so as to determine the partial component group for each comparison target image. A comparison target image having high consistency with one of the sample images is extracted by comparison and collation between these partial component groups and the partial component groups which include each element of each partial normalized orthogonal system among component groups of each sample image. And a new comparison target image group is selected near the extracted image, and a new partial normalized orthogonal system is created by increasing the dimensions. Then (2) is repeated.

Image processor

The entire region of variation in the viewing range of the attitude of a workpiece is divided roughly, and images of the workpiece are captured from each direction. The images are stored together with imaging direction data as a first teaching model. Images captured by varying the imaging direction at a narrow pitch within a predetermined range of attitude variation in the workpiece are then stored together with imaging direction data as a second teaching model. Images of the workpiece are captured and compared with the teaching models. The position and attitude of the workpiece are determined by the imaging direction and imaging position of the selected teaching model.

A group of bolts are suppled into a placing surface of a tray, an isolated bolt is searched for by a visual sensor, and its deviation from a standard position at the time of teaching is determined. The isolated bolt is picked up by a robot that has been taught how to pick up an isolated bolt laying in a standard position. The position of the robot's hand is corrected according to the deviation from the standard position before the robot attempts to pick up the located isolated bolt. If no isolated bolt is found, a shaking device,is operated to loosen the piled-up bolts, and a new isolated bolt is again searched for by the visual sensor. The isolated bolt, if found, is picked up. The picking-up operation may be performed by searching for an isolated small set of bolts using a three-dimensional visual sensor. The oscillating excitation can also be provided by a robot. Also, the piled-up bolts may be loosened or disentangled using a robot equipped with a tool for leveling the pile of bolts or by using a gas jet.

Apparatus for picking up articles using a vision controlled robot

Curve interpolation method for performing velocity control during

A method of tracking a robot with respect to a circularly moving workpiece W. When a workpiece on a disc-shape conveyer remains stationary at a reference position W0, positions P0 and Q0 are taught to the robot in a stationary coordinate system .SIGMA.0 and the robot is placed on stand-by at a position A. The angular displacement of the disc-shape conveyer is detected by a pulse encoder and counting of the output pulses starts when the workpiece W arrives at the reference position W0. A CPU of a robot controller reads the counted amount in a short cycle and transforms it into a rotation amount .theta. of the conveyer from the reference position. Updating of matrix data for setting a rotary coordinate system .SIGMA.rot based on the rotation amount .theta. is repeated. When the workpiece W reaches the reference position W0, the CPU executes interpolation calculation for successively determining a target point to which the robot is moved, based on the position data of the waiting position and the position data of teaching points P0 and Q0. The position data obtained by the interpolation calculation in the rotary coordinate system .SIGMA.rot is converted into the position data in the stationary coordinate system .SIGMA.0 and the position of the robot is controlled based on thus obtained position data.

Growing community of inventors

Yamanashi, Japan

Hidetoshi Kumiya