WO2019176478A1 - ロボットの動作制御装置 - Google Patents
ロボットの動作制御装置 Download PDFInfo
- Publication number
- WO2019176478A1 WO2019176478A1 PCT/JP2019/006183 JP2019006183W WO2019176478A1 WO 2019176478 A1 WO2019176478 A1 WO 2019176478A1 JP 2019006183 W JP2019006183 W JP 2019006183W WO 2019176478 A1 WO2019176478 A1 WO 2019176478A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- processing
- evaluation value
- robot
- evaluation
- candidate
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- 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
- B25J9/1628—Programme controls characterised by the control loop
- B25J9/163—Programme controls characterised by the control loop learning, adaptive, model based, rule based expert control
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J13/00—Controls for manipulators
- B25J13/08—Controls for manipulators by means of sensing devices, e.g. viewing or touching devices
-
- 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
- B25J9/1628—Programme controls characterised by the control loop
- B25J9/1635—Programme controls characterised by the control loop flexible-arm control
-
- 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
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/008—Artificial life, i.e. computing arrangements simulating life based on physical entities controlled by simulated intelligence so as to replicate intelligent life forms, e.g. based on robots replicating pets or humans in their appearance or behaviour
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- 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
- B25J9/1602—Programme controls characterised by the control system, structure, architecture
- B25J9/161—Hardware, e.g. neural networks, fuzzy logic, interfaces, processor
-
- 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
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/39—Robotics, robotics to robotics hand
- G05B2219/39311—Multilayer, MNN, four layer perceptron, sigmoidal neural network
-
- 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
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/40—Robotics, robotics mapping to robotics vision
- G05B2219/40053—Pick 3-D object from pile of objects
Definitions
- the present invention relates to a robot motion control device, a motion control method, a motion control program, and a processing device.
- Patent Document 1 discloses a control device that controls a robot arm based on a shape model of the object included in image data in the control of gripping the object by the robot arm.
- This document also discloses a learning device that outputs a gripping success probability using image data as input in order to control a robot arm.
- Such a problem is not limited to the robot motion control, and may be a problem that may occur in all processing apparatuses that perform output for a predetermined process by a learning device, for example.
- the present invention has been made to solve this problem, and in the case of learning additional processing, a robot motion control device, a motion control method, a motion control program, and a robot that can efficiently perform learning, and
- An object is to provide a processing apparatus.
- the present disclosure is a robot motion control apparatus for a processing object, at least a captured image obtained by capturing the processing object, an input unit to which at least one motion candidate of the robot is input, the captured image, and In order to output at least one of the motion candidates as an input and output a first evaluation value indicating the evaluation of each motion candidate when the robot performs a first processing motion, learning is performed using the first learning data.
- the second evaluation value indicating the evaluation of each motion candidate when the robot performs the second processing operation using the learned first learner, the captured image and at least one motion candidate as inputs.
- the second learning data learned by the second learning data different from the first learning data and at least one of the first evaluation value and the second evaluation value Based on the people, and a, and evaluation unit which calculates a command value for operating the robot.
- an additional learning device that considers only that processing operation is generated. is doing. That is, in the present invention, in order to perform a second processing operation different from the first processing operation, a second learning device that has been learned is separately generated using learning data different from the first processing operation, and the first In addition, the operation command is calculated in the evaluation unit in consideration of the output of the second learning device. Therefore, the learning burden on the learning device can be reduced.
- the processing object is an object such as a part to be worked on by the robot and an object to be transported
- the processing operation is an operation performed by the robot on the processing object.
- the motion candidate is a motion candidate that the robot performs when the robot performs a processing operation.
- the robot hand moves forward and backward, moves up and down, and rotates.
- two rotational motions with different rotation angles can be motion candidates, even for the same motion type.
- the first evaluation value can be a probability value indicating a success probability of the first motion candidate
- the second evaluation value is a success probability of the second motion candidate.
- the evaluation unit can calculate the command value based on motion candidates in which each evaluation value is a predetermined value or more.
- the evaluation unit can calculate the command value by weighting the first evaluation value and the second evaluation value based on each processing operation.
- the second evaluation value may be an evaluation value of an event that may occur due to at least one of the first processing operation and the second processing operation.
- a third evaluation value indicating an event that can occur due to at least one of the first processing operation and the second processing operation, with the captured image and at least one of the operation candidates as inputs.
- the third learning device that has been learned with the third learning data different from the first and second learning data can be further provided, and the evaluation unit includes the third learning data.
- the command value can also be calculated using the evaluation value.
- the captured image is input, and the first operation candidate suitable for the first processing operation and the second operation candidate suitable for the second processing operation are output.
- An operation candidate generation unit can be further provided.
- the operation candidates for the first processing operation and the second processing operation may include a movement vector for operating the robot.
- the present disclosure is a robot control system, comprising a robot that performs processing on a processing object and any one of the motion control devices described above, and according to a command value calculated from the motion control device,
- the robot is configured to operate.
- the present disclosure is a method for controlling the operation of a robot with respect to a processing object, the step of inputting at least a captured image obtained by capturing the processing object and at least one motion candidate of the robot, the captured image, and at least one Learning that has been learned from the first learning data in order to output the first evaluation value indicating the evaluation of each of the motion candidates when the robot performs the first processing operation.
- Preparing a learned second learned device that has been learned using second learning data different from the first learning data, and outputting the evaluation value of A step of outputting the first evaluation value from the image and the motion candidate by the first learner; and a step of outputting the second evaluation value from the captured image and the motion candidate by the second learner. And a step of calculating a command value for operating the robot based on at least one of the first evaluation value and the second evaluation value.
- the present disclosure is a robot motion control program for a processing object, wherein at least a captured image obtained by capturing the processing object and at least one motion candidate for the robot are input to a computer; In order to output a first evaluation value indicating an evaluation of each motion candidate when the robot performs a first processing operation with at least one of the motion candidates as input, learning is performed using first learning data. Preparing the learned first learning device, and taking the captured image and at least one motion candidate as inputs, and evaluating each motion candidate when the robot performs a second processing operation. In order to output a second evaluation value, a learned second learner that has been learned using second learning data different from the first learning data is prepared.
- the present disclosure is a control apparatus for a control object, an input unit that inputs environment information related to the control object and at least one processing candidate for the control object, the environment information, and at least one
- learning is performed by the first learning data.
- the second learning indicating the evaluation of each of the processing candidates when the second processing is performed on the control target object by inputting the learned first learning device, the environment information, and at least one of the processing candidates.
- the evaluation value of the second learning data learned from the second learning data different from the first learning data the first evaluation value, and the second evaluation value Based on at least one
- a, and evaluation unit which calculates a command value for controlling the controlled object.
- the present disclosure is a processing device that performs a predetermined process, and includes an input unit that inputs at least one processing candidate, and at least one processing candidate as an input.
- the first evaluation value indicating the evaluation the first learning data learned by the first learning data and at least one of the processing candidates are input and the second processing is performed.
- a second evaluation value indicating the evaluation of each processing candidate a learned second learner learned by second learning data different from the first learning data, and the second An evaluation unit that determines at least one process candidate of the first process and the second process based on at least one of the first evaluation value and the second evaluation value.
- the learning when an additional process is learned, the learning can be performed efficiently.
- FIG. 1 It is the schematic which shows an example of the robot used with one Embodiment of the operation control apparatus of the robot of this invention. It is a functional block diagram of one embodiment of an operation control device concerning the present invention. It is a figure which shows holding
- FIG. 3 is a flowchart illustrating an example of a processing procedure for updating a learning device in the motor control system of FIG. 2. It is a functional block diagram of one embodiment of a control device concerning the present invention. It is a functional block diagram of one embodiment of a processing unit concerning the present invention.
- FIG. 1 is a schematic diagram of a robot according to the present embodiment
- FIG. 2 is a block diagram showing a functional configuration of a robot control system including an operation control device.
- This control system is a system for controlling the operation of the robot, and gives an operation command for appropriately processing the object to be processed by the robot.
- FIG. 1 a case will be described as an example where the robot 1 having the robot hand 10 issues an operation command for gripping a target component 50 among a plurality of components (processing objects) 5.
- the operation control device 2 operates as shown in FIG. 2 in which the robot hand 10 operates to hold the captured image of the robot hand 10 and the target component 50 and the target component 50.
- a plurality of motion candidates to be input are input to the learning devices 211 to 213. That is, the same inputs are made to the learning devices 211 to 213.
- three learning devices 211 to 213 are provided.
- the first learning device 211 will be described first.
- the first learning device 211 is trained so that when a captured image and a plurality of motion candidates are input, the success probability that the target part of each motion candidate can be grasped is output as a first probability value.
- the captured image is an image of the robot hand 10 and the target component 50, and is for extracting the current positions of the robot hand 10 and the target component 50. However, when the robot hand 10 is arranged at the fixed initial position, a captured image obtained by capturing only the target component 50 can be prepared.
- a movement vector Mn (xn, yn, ⁇ n) (n is an integer) for moving the robot hand 10 toward the target component 50 is used as a motion candidate.
- x represents the amount of movement in the x direction
- y represents the amount of movement in the y direction
- ⁇ represents the tilt of the robot hand 10.
- the motion candidate showing the highest first probability value can be selected.
- an operation command is not performed and an operation candidate different from the operation candidate used for input is prepared, and then the output of the probability value is renewed by the first learning device 211. It can also be done.
- Such a first learning device 211 can be configured by machine learning such as a neural network.
- the motion hand is selected for the purpose of gripping the target component 50 by the robot hand 10, but a further condition is added to cause the robot hand 10 to grip the target component 50.
- a further condition is added to cause the robot hand 10 to grip the target component 50.
- the robot hand 10 may grip the center X of the target component 50. You may want to add it. In this case, it is necessary to relearn the first learning device 211 so that the success probability that the robot hand 10 reaches the center of the target component 50 can be output.
- the robot hand 10 according to the additional condition is operated by the second learning device 212 different from the first learning device 211, and each operation candidate succeeds under the added condition. A second probability value indicating the probability is output.
- the evaluation unit 214 considers two of the first probability value of the motion candidate selected by the first learner 211 and the second probability value of the motion candidate output by the second learner 212.
- the command is calculated. That is, the first learning device 211 is not trained to consider all conditions, but when a condition is added, a second learning device 212 that additionally considers only the condition is additionally generated, and the first learning is performed.
- the evaluation unit 214 calculates an operation command in consideration of the outputs of both the device 211 and the second learning device 212. Therefore, the learning burden on the learning device can be reduced.
- the operation control device 2 of the present embodiment includes a third learning device 213.
- the third learning device 213 can output a third probability value indicating the probability that the target component 50 can be gripped without affecting other components when the target component 50 is gripped by the robot hand 10.
- the robot hand 10 can output the probability that the target component 50 can be gripped without changing the positions of the other components 51 to 53 shown in FIG.
- the first and second learners 211 and 312 generate a third learner 213 that performs learning separately, and the evaluation unit 214 outputs the third probability value that is the output.
- the operation command is calculated in consideration of the first and second probability values.
- the number of learning devices is not particularly limited, and a new learning device can be generated each time an additional condition is added.
- the robot when control for causing the robot 10 to perform a predetermined target motion is performed, instead of learning one learner so as to satisfy all the conditions, the robot should perform the motion to be performed.
- the conditions are divided, the operation of the robot 10 is learned for each condition, and the operation is evaluated by the learning device for each condition.
- the motion command to be performed by the robot 10 is calculated by comprehensively considering the motion evaluation output by each learning device.
- each evaluation can be considered equally or weighted.
- a priority order of evaluation may be provided, and for example, when one evaluation is not a predetermined evaluation or higher, other evaluations may not be considered. Thereby, the malfunction at the time of using one learning device as mentioned above can be eliminated.
- FIG. 4 is a block diagram showing a hardware configuration of the robot control system according to the present embodiment.
- the robot 1 has a robot hand 10 that can grip a component.
- the robot hand 10 moves up and down, left and right, and the robot hand 10 tilts.
- the configurations of the robot 1 and the robot hand 10 are not particularly limited, and can be appropriately changed so that other operations can be performed.
- the robot motion control apparatus 2 includes a control unit 21, a storage unit 22, a communication interface 23, an input device 24, an output device 25, a display device 26, an external interface 27, and a drive.
- Reference numeral 28 denotes an electrically connected computer.
- the communication interface and the external interface are described as “communication I / F” and “external I / F”, respectively.
- the control unit 21 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), and the like, and controls each component according to information processing.
- the storage unit 22 is, for example, an auxiliary storage device such as a hard disk drive or a solid state drive, and includes a control program 221 executed by the control unit 21, operation candidate data 222, photographed image data 223, probability value data 224, first to Learning result data 225 indicating information related to the third learning devices 211 to 213, learning data 226 for learning the first to third learning devices 211 to 213, and the like are stored. In addition, various data necessary for driving the operation control device 2 can also be stored.
- the control program 221 receives the above-described input for the operation of the robot 1 and performs output according to the input.
- the control program 221 performs output by first to third learning devices 211 to 213 described later.
- the motion candidate data 222 is data including a movement vector of the robot hand 10 and has a large number of motion candidates.
- the motion candidate data is prepared in advance according to the type of the robot 1 and the type of the target article.
- the photographed image data 223 is data related to the photographed image obtained by photographing the robot hand 10 and the target component 50 as described above, and the probability value data 224 is each operation of the robot hand 10 output by the learning devices 211 to 213. It is data which shows the success probability of a candidate.
- the learning result data 225 is data for setting the learning devices 211 to 213 that have already been learned.
- the learning data 226 is data used for the learning of each of the current learning devices 211 to 213. A detailed description regarding learning will be described later.
- the communication interface 23 is, for example, a wired LAN (Local Area Network) module, a wireless LAN module, or the like, and is an interface for performing wired or wireless communication via a network. For example, it is possible to communicate with the robot 1 to transmit an operation command, or to communicate with an external camera 4 and acquire a photographed image photographed by the camera 4. It can also be used to transmit information related to the control of the robot 1 to the outside.
- the input device 24 is a device for performing input such as a mouse and a keyboard, for example, and can input various instructions related to the operation of the robot 1 by the user.
- the output device 25 is a device for outputting, for example, a speaker.
- the display device 26 can be composed of a display or the like, and can display the output results from the learning devices 211 to 213, for example.
- the external interface 27 is a USB (Universal Serial Bus) port or the like, and is an interface for connecting to an external device.
- the drive 28 is, for example, a CD (Compact Disk) drive, a DVD (Digital Versatile Disk) drive, or the like, and is a device for reading a program stored in the storage medium 91.
- the type of the drive 28 may be appropriately selected according to the type of the storage medium 91.
- At least one of various data 221 to 226 stored in the storage unit 22 may be stored in the storage medium 91.
- the captured image data 223 and the probability value data 224 can be stored in the RAM of the control unit 21.
- the storage medium 91 stores information such as a program by an electrical, magnetic, optical, mechanical, or chemical action so that information such as a program recorded by a computer or other device or machine can be read. It is a medium to do.
- the operation control apparatus 2 may acquire the various data 221 to 226 from the storage medium 91.
- a disk-type storage medium such as a CD or a DVD is illustrated.
- the type of the storage medium 91 is not limited to the disk type and may be other than the disk type.
- Examples of the storage medium other than the disk type include a semiconductor memory such as a flash memory.
- control unit 21 may include a plurality of processors.
- the operation control device 2 may be configured by a plurality of information processing devices.
- the operation control device 2 may be a general-purpose desktop PC (Personal Computer), a tablet PC, or the like, in addition to an information processing device designed exclusively for the provided service.
- control unit 21 of the operation control device 2 expands the control program 221 stored in the storage unit 22 in the RAM. Then, the control unit 21 interprets and executes the program 221 expanded in the RAM, and controls each component. As a result, as shown in FIG. 2, the control device 2 according to the present embodiment functions as a computer including an input unit 210, first to third learning devices 211 to 213, and an evaluation unit 214.
- the input unit 210 prepares captured images and motion candidates as inputs to the learning devices 211 to 213. For example, when the control of the robot 1 is started by an instruction from the input device 24, the camera 4 is caused to take an image of the robot hand 10 and the target component 20 to obtain a taken image.
- the captured image can be stored in the storage unit 22 as captured image data 223.
- a necessary captured image can be read from the captured image data in the storage unit 22.
- a plurality of motion candidates are selected for input from the motion candidate data 222 of the storage unit 22. At this time, the motion candidate for input is selected at random from the motion candidate data 222. In this way, the captured image and the plurality of motion candidates prepared by the input unit 210 are input to the learning devices 211 to 213.
- the first learning device 211 is trained to receive a captured image and a plurality of motion candidates as inputs, and to output a first probability value indicating a success probability that the robot hand 10 can grip the target component 50 by the motion of each motion candidate. Yes.
- the operation in which the robot hand 10 simply holds the target component 50 is referred to as a first processing operation.
- the first learning device 211 that performs such an output is composed of a neural network. Specifically, it is a neural network having a multilayer structure used for so-called deep learning as shown in FIG. 5, and includes an input layer 71, an intermediate layer (hidden layer) 72, and an output layer 73 in order from the input. .
- the neural network 7 includes one intermediate layer 72, and the output of the input layer 71 is an input of the intermediate layer 72, and the output of the intermediate layer 72 is an input of the output layer 73.
- the number of intermediate layers 72 is not limited to one, and the neural network 7 may include two or more intermediate layers 72.
- Each layer 71 to 73 includes one or a plurality of neurons.
- the number of neurons in the input layer 71 can be set according to the number of input data.
- the number of neurons in the intermediate layer 72 can be set as appropriate according to the embodiment.
- the output layer 73 can also be set according to the number of motion candidates.
- Adjacent layers of neurons are appropriately connected to each other, and a weight (connection load) is set for each connection.
- each neuron is connected to all neurons in the adjacent layers, but the neuron connection is not limited to such an example, and is appropriately set according to the embodiment. It's okay.
- a threshold is set for each neuron, and basically, the output of each neuron is determined by whether or not the sum of products of each input and each weight exceeds the threshold.
- the motion control apparatus 2 obtains an output indicating the success probability of each motion candidate from the output layer 73 by performing the above input to the input layer 71 of the neural network 7.
- the configuration of the neural network 7 (for example, the number of layers of the neural network 7, the number of neurons in each layer, the connection relationship between neurons, the transfer function of each neuron), the weight of connection between each neuron, and each neuron
- the information indicating the threshold value is included in the learning result data 225.
- the motion control device 2 refers to the learning result data 225 and sets a learned learner.
- learning by the first learning device 211 is performed by learning data 226 as shown in FIG. 6, for example.
- the learning image 226 is input using the captured image and the motion candidate as input and the success or failure of gripping by the robot hand 10 as an output, and learning is performed by the error propagation method or the like.
- An example of specific learning data 226 is shown in FIG.
- the learning data is obtained by combining the photographed image and the motion candidate in various ways and actually driving the robot hand 10 to determine whether the gripping has succeeded or not. That is, when a captured image in which the robot hand 10 and the target component 50 are captured at a predetermined position is acquired and the robot hand 10 is operated with a predetermined motion candidate for this, the robot hand 10 actually grips the target component 50.
- the learning data 226 is generated as success when it is successful and as failure when it cannot be grasped.
- FIG. 7 includes learning data indicating that the target component 20 has been gripped when the robot hand 10 is driven based on the image 1 and the motion candidate M1.
- a success probability that can grasp the target component 50 that is, a first probability value can be output in each motion candidate. That is, learning result data 225 for the first learning device 211 that can obtain such an output is generated.
- the learning data 226 can be created by actually using the robot hand 10, or the learning data 226 is created by operating the robot hand by simulation on a computer. You can also.
- the second learning device 212 receives a captured image and a plurality of motion candidates as input, and learns to output a second probability value indicating the success probability that the robot hand 10 can reach the center of the target component 50 by the motion of each motion candidate. Has been.
- a second processing operation an operation in which the robot hand 10 reaches the center of the target component 50 is referred to as a second processing operation.
- the second learning device 212 can also be configured by a neural network, and thus detailed description thereof is omitted.
- the learning method and the generation of learning data are also substantially the same as those of the first learning device 211.
- Third learner> In the third learning device 213, a captured image and a plurality of motion candidates are input, and the probability that the other components 51 to 53 do not move in the process until the robot hand 10 reaches the target component 50 by the motion of each motion candidate. Learning to output the third probability value shown. That is, the probability that the other parts 51 to 53 do not move without being influenced by the robot hand 10 or the target part 50 is output.
- a third processing operation an operation that does not move other components is referred to as a third processing operation.
- the third learning device 213 can also be configured by a neural network like the first learning device 211, detailed description thereof is omitted.
- the learning data 226 can be generated as follows, for example.
- the learning data 226 is obtained by actually driving the robot hand 10 to determine whether or not the other components 51 to 53 have moved in response to various combinations of captured images and motion candidates. For example, when a captured image in which the robot hand 10 and the target component 50 are captured at a predetermined position is acquired and the robot hand 10 is moved to the target component 50 with a predetermined operation candidate, the other components do not move. 1 if it has moved, 0 if it has moved.
- FIG. 8 shows an example of learning data.
- the number of pixels that have changed in the images before and after the operation of the robot hand 10 is greater than or equal to a predetermined reference, it can be determined that the components 51 to 53 have moved. Further, instead of using the probability that another component operates as an output, the number of changed pixels before and after the operation can be output.
- the captured image is divided into a plurality of ranges, and the change in the number of pixels before and after the operation is calculated for each divided range. Then, the third learning device 213 can be made to learn so that the influence of the movement of the component is output in the range of 0 to 1 based on the change in the number of pixels for each divided range.
- the learning data 226 thus obtained is used for learning by the third learning device 213 as in the first embodiment.
- learning result data 225 for the third learning device 213 that can obtain the output is generated.
- the evaluation unit 214 determines an operation command for the robot 1 based on the probability values of the operation candidates output from the learning devices 211 to 213. As shown in FIG. 9, the probability values p 1 to p 3 of the motion candidates M 1 to Mn are output from the learning devices 211 to 213 to the evaluation unit 214.
- the evaluation values P of the motion candidates M1 to Mn are calculated, and the motion candidate having the highest evaluation value P is output to the robot 1 as a motion command.
- the success probability shown in FIG. 9 is output, the sum of the success probabilities p1 to p3 is set as the evaluation value P, and the motion candidate having the highest evaluation value P is set as the motion command to the robot 1. Can do.
- the success probabilities p1 to p3 can be weighted. For example, in consideration of the priority of the first to third processing operations, the weight of each processing operation is set to w1, w2, and w3.
- the evaluation value P of each motion candidate can be expressed as w1 * p1 + w2 * p2 + w3 * p3, and the motion candidate that has the highest evaluation value can be used as the motion command.
- the output weight w2 is smaller than the weight w1.
- the evaluation value P can be calculated by weighting. For example, when the third learning device 213 outputs the number of pixels that have changed as described above, the output from the first and second learning devices 211 and 212 is obtained by performing such normalization. Can be adapted to the range.
- the motion candidate may not be adopted as the motion command.
- FIG. 10 is a flowchart illustrating an example of a processing procedure of the operation control apparatus. Note that the processing procedure described below is merely an example, and each processing may be changed as much as possible. Further, in the processing procedure described below, steps can be omitted, replaced, and added as appropriate according to the embodiment.
- a captured image is acquired by the camera 4 (step S101).
- the acquired photographed image and motion candidate are input to the learning devices 211 to 213 (step S102).
- motion candidates and their probability values are output from the learning devices 211 to 213 (step S103).
- the evaluation unit 214 generates an operation command based on the output operation candidate and the probability value (step S104).
- the generated operation command is transmitted to the robot 1, and the robot hand 10 is driven according to the operation command (step S105).
- one learner is not trained to consider all conditions, but when a condition is added, a learner that considers only the condition is added.
- the evaluation unit 214 calculates an operation command in consideration of the generated outputs of all the learning devices 211 to 213. Therefore, the learning burden on the learning device can be reduced.
- the number of learning devices is three.
- the number of learning devices may be two or more, and is not particularly limited.
- the processing operation that is the target of the first learning device 211 and the processing operations of the second and third learning devices 212 and 213 may be related to each other.
- processing operations that restrict the first processing operation such as the first processing operation and the second processing operation of the above-described embodiment, may be related to become the second processing operation.
- the processing operation related to the event that may occur due to the first processing operation such as the first processing operation and the third processing operation, may be related to become the third processing operation.
- the first to third learning devices 211 to 213 are provided.
- the motion control device 2 may be configured to include the first learning device 211 and the third learning device 213. .
- each of the learning devices 211 to 213 outputs the success probability of the motion candidate as an evaluation value, but is not limited to this. That is, as the evaluation value of each motion candidate, in addition to the success probability, for example, the coordinate of the movement destination of the robot hand 10 based on each motion candidate can be used, and the motion candidate can be evaluated using this coordinate. . For example, if the coordinates of the movement destination are close to the coordinates of the target component 50, the evaluation value is high. In addition to this, it is possible to evaluate motion candidates using various evaluation values and output them from the learning devices 211 to 213.
- the motion candidates input to the learning devices 211 to 213 include various selection methods other than the random selection as described above.
- the motion control device 2 is provided with a motion candidate generation unit that generates motion candidates.
- this motion candidate generation unit first, a predetermined average and variance of evaluation values are set, and based on these, a certain number of motion candidates are sampled to generate a set of motion candidates.
- the evaluation value of each motion candidate in the set is output and ranked in descending order of the evaluation value.
- several motion candidates with high rank in the set are selected, and the average and variance of the evaluation values of the motion candidates are calculated.
- the average and variance of the set are updated based on the average and variance. By repeating this, an operation candidate having a high evaluation value can be generated.
- the motion candidate generation unit prepares in advance a set of motion candidates suitable for each motion of the robot hand 10, and from among them, the motion candidate is selected according to the requested motion of the robot hand 10.
- the set can be selected as appropriate.
- each neural network 7 may be a convolutional neural network that uses the input layer 71 and the intermediate layer 72 as a convolution layer and a pooling layer.
- each neural network 7 may be a recursive neural network having a connection that recurs from the output side to the input side, such as the intermediate layer 72 to the input layer 71. Note that the number of layers of each neural network 7, the number of neurons in each layer, the connection relationship between neurons, and the transfer function of each neuron may be appropriately determined according to the embodiment.
- each of the learning devices 211 to 213 is not particularly limited.
- a support vector machine, a self-organizing map, or a learning device that performs learning by reinforcement learning can be used.
- other methods other than machine learning can be used.
- the robot motion control device 2 has been described. However, as in this device, each time an additional condition is added, a new learner is generated and the outputs of a plurality of learners are evaluated. Can also be applied to other devices.
- the present invention can be applied to a control device 8 as shown in FIG.
- the basic configuration of the control device 8 is substantially the same as that of the robot motion control device 2 described above.
- the information input to the learning device is not limited to an image, and various information referred to as environment information can be input.
- the environmental information is information related to the control object, and can be, for example, the operation speed of the control object, the position of the control object, etc. in addition to the image of the control object.
- the processing candidate corresponds to the motion candidate, and is a candidate for processing (control) on the control target.
- this control device can be applied to control for collision avoidance of automobiles.
- input environmental information includes measured values such as distances to obstacles obtained by measuring instruments such as cameras and radars installed in automobiles, and multiple driving operations with different steering and accelerator operation amounts as processing candidates. Can be used.
- the first learning device 211 can output the collision probability with the obstacle from these inputs. Therefore, by using the first learning device 211, it is possible to determine a driving operation that minimizes the collision probability of the autonomous driving vehicle.
- the evaluation unit 214 can determine a driving operation with a small burden on the occupant while avoiding a collision.
- the target of the first learning device 211 is a collision, it can be said to be an indispensable condition, and the target of the second learning device 212 can be referred to as a recommended condition having a lower priority. Therefore, the output weight of the first learning device 211 is larger than the output weight of the second learning device 212. Three or more learning devices may be provided.
- the present invention can be applied not only to control of the control object but also to processing of the process object.
- the present invention can be applied to a processing apparatus as shown in FIG.
- the basic configuration of the processing device 9 is substantially the same as that of the control device 8 described above.
- the information input to the learning device can be only processing candidates.
- Such a processing apparatus 9 can be applied to an image search system, for example.
- an image is input as a processing candidate, and the first learning device 211 can output the probability that a person is reflected in the input image. Thereby, it is possible to search for an image showing a person.
- An image is also input as a processing candidate to the second learning device 212.
- the second learning device 212 can output a probability that a person shown in the image belongs to a certain age.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Robotics (AREA)
- Mechanical Engineering (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Manipulator (AREA)
Abstract
Description
以下、本実施形態に係るロボットの動作制御装置について図1及び図2を参照しつつ説明する。図1は、本実施形態に係るロボットの概略図、図2は、動作制御装置を含むロボットの制御システムの機能構成を示すブロック図である。
<2-1.ハードウエア構成>
次に、本実施形態に係るロボットの動作制御装置2を含む制御システムのハードウエア構成について、図4も参照しつつ説明する。図4は、本実施形態に係るロボットの制御システムのハードウエア構成を示すブロック図である。
図1に示すように、このロボット1は、部品を把持可能なロボットハンド10を有しており、このロボットハンド10が上下左右に移動するほか、ロボットハンド10が傾くように構成されている。但し、ロボット1及びロボットハンド10の構成は特には限定されず、他の動作を行えるように適宜変更可能である。
図4に示すように、本実施形態に係るロボットの動作制御装置2は、制御部21、記憶部22、通信インタフェース23、入力装置24、出力装置25、表示装置26、外部インタフェース27、及びドライブ28が電気的に接続されたコンピュータである。なお、図4では、通信インタフェース及び外部インタフェースをそれぞれ、「通信I/F」及び「外部I/F」と記載している。
次に、図2及び図4を参照しつつ、本実施形態に係る動作制御装置の機能構成の一例を説明する。
図2に示すように、動作制御装置2の制御部21は、記憶部22に記憶された制御プログラム221をRAMに展開する。そして、制御部21は、RAMに展開されたプログラム221をCPUにより解釈及び実行して、各構成要素を制御する。これによって、図2に示すように、本実施形態に係る制御装置2は、入力部210、第1~第3学習器211~213、及び評価部214を備えるコンピュータとして機能する。
入力部210は、各学習器211~213への入力として、撮影画像及び動作候補を準備する。例えば、入力装置24からの指示により、ロボット1の制御が始まると、カメラ4にロボットハンド10及び対象部品20を撮影させ、撮影画像を取得する。撮影画像は、撮影画像データ223として記憶部22に記憶することができる。なお、予め撮影画像が取得されている場合には、記憶部22の撮影画像データから必要な撮影画像を読み出すこともできる。また、記憶部22の動作候補データ222から複数の動作候補を入力用に選択する。このとき、入力用の動作候補は、動作候補データ222の中からランダムに選択する。こうして、入力部210により準備された撮影画像及び複数の動作候補を各学習器211~213に入力する。
第1学習器211では、撮影画像及び複数の動作候補を入力とし、各動作候補による動作によってロボットハンド10が対象部品50を把持できる成功確率を示す第1確率値を出力するように学習されている。ここで、ロボットハンド10が対象部品50を単に把持する動作を第1処理動作と称することとする。
第2学習器212では、撮影画像及び複数の動作候補を入力とし、各動作候補による動作によってロボットハンド10が対象部品50の中心に到達できる成功確率を示す第2確率値を出力するように学習されている。ここで、ロボットハンド10が対象部品50の中心に到達するような動作を第2処理動作と称することとする。
第3学習器213では、撮影画像及び複数の動作候補を入力とし、各動作候補による動作によってロボットハンド10が対象部品50に到達するまでの過程で、他の部品51~53が移動しない確率を示す第3確率値を出力するように学習されている。すなわち、ロボットハンド10によって、あるいは対象部品50によって、他の部品51~53が影響を受けずに移動しない確率を出力する。ここで、ロボットハンド10が対象部品50に到達する過程で、他の部品を移動させないような動作を第3処理動作と称することとする。
評価部214は、各学習器211~213から出力された、各動作候補の確率値に基づいて、ロボット1に対する動作指令を決定するものである。各学習器211~213からは、図9に示すように、各動作候補M1~Mnの確率値p1~p3が評価部214へ出力される。
次に、図10を参照しつつ、動作制御装置2におけるロボットの制御処理の例を説明する。図10は、動作制御装置の処理手順の一例を例示するフローチャートである。なお、以下で説明する処理手順は一例に過ぎず、各処理は可能な限り変更されてよい。また、以下で説明する処理手順について、実施の形態に応じて、適宜、ステップの省略、置換、及び追加が可能である。
以上のように、本実施形態によれば、一つの学習器を、すべての条件を考慮するように学習させるのではなく、条件が付加されたときには、その条件のみを考慮した学習器を追加で生成し、生成したすべての学習器211~213の出力を考慮して、評価部214において、動作指令を算出するようにしている。したがって、学習器の学習の負担を低減すことができる。
以上、本発明の実施の形態を詳細に説明してきたが、前述までの説明はあらゆる点において本発明の例示に過ぎない。本発明の範囲を逸脱することなく種々の改良や変形を行うことができることは言うまでもない。例えば、以下のような変更が可能である。なお、以下では、上記実施形態と同様の構成要素に関しては同様の符号を用い、上記実施形態と同様の点については、適宜説明を省略した。以下の変形例は適宜組み合わせ可能である。
上記実施形態では、学習器の数が3つであるが、学習器の数は、2以上あればよく、特には限定されない。また、第1学習器211の対象となる処理動作と、第2,第3学習器212,213の処理動作は、関連していればよい。例えば、上記実施形態の第1処理動作と第2処理動作のように、第1処理動作を制約するような処理動作が、第2処理動作となるように関連していてもよい。また、第1処理動作と第3処理動作のように、第1処理動作に起因して生じうる事象に関連する処理動作が第3処理動作となるように関連していてもよい。
上記実施形態では、第1~第3学習器211~213を設けたが、例えば、動作制御装置2が、第1学習器211と第3学習器213とを有するように構成されていてもよい。
上記実施形態では、各学習器211~213において、動作候補の成功確率を評価値として出力しているが、これに限定されない。すなわち、各動作候補の評価値としては、成功確率以外に、例えば、各動作候補に基づくロボットハンド10の移動先の座標とすることもでき、この座標によって、動作候補の評価を行うことができる。例えば、移動先の座標が対象部品50の座標に近ければ、評価値が高くなる。これ以外にも、種々の評価値を用いて、動作候補の評価を行い、これを学習器211~213から出力することができる。
各学習器211~213に入力する動作候補は、上記のようにランダムに選択する以外に種々の選択方法がある。例えば、動作制御装置2に、動作候補を生成する動作候補生成部を設ける。この動作候補生成部では、まず、評価値の所定の平均と分散を設定し、これらに基づいて、一定数の動作候補をサンプリングして、動作候補の集合を生成する。次に、学習器を用いて、集合内の各動作候補の評価値を出力し、評価値の高い順にランク付けを行う。続いて、集合内のランクの高い動作候補をいくつか選択し、その動作候補の評価値の平均と分散を算出する。そして、その平均と分散により、集合の平均と分散を更新する。これを繰り返すことで、評価値の高い動作候補を生成することができる。
上記の例では、ニューラルネットワーク7として、多層構造を有する一般的な順伝播型ニューラルネットワークを用いている。しかしながら、各ニューラルネットワーク7の種類は、このような例に限定されなくてもよく、実施の形態に応じて適宜選択されてよい。例えば、各ニューラルネットワーク7は、入力層71及び中間層72を畳み込み層及びプーリング層として利用する畳み込みニューラルネットワークであってもよい。また、例えば、各ニューラルネットワーク7は、中間層72から入力層71等のように出力側から入力側に再帰する結合を有する再帰型ニューラルネットワークであってもよい。なお、各ニューラルネットワーク7の層数、各層におけるニューロンの個数、ニューロン同士の結合関係、及び各ニューロンの伝達関数は、実施の形態に応じて適宜決定されてよい。
上記各学習器211~213の種類は、特には限定されず、ニューラルネットワーク以外に、例えば、サポートベクターマシン、自己組織化マップ、又は強化学習により学習を行う学習器とすることができる。あるいは、機械学習以外の他の手法を用いることができる。
上記実施形態では、ロボットの動作制御装置2について説明したが、この装置のように、付加的な条件が加えられるたびに、新たな学習器を生成し、複数の学習器の出力を評価する手法は、他の装置にも適用することができる。
また、制御対象物の制御だけでなく、処理対象物に対する処理にも適用することができる。例えば、図12に示すような処理装置に適用することができる。この処理装置9の基本構成は、上述した制御装置8と概ね同じである。但し、学習器に入力される情報は、処理候補だけにすることができる。
212 第2学習器
213 第3学習器
214 評価部
Claims (13)
- 処理対象物に対するロボットの動作制御装置であって、
少なくとも前記処理対象物を撮影した撮影画像、及び前記ロボットの少なくとも一つの動作候補が入力される入力部と、
前記撮影画像、及び少なくとも一つの前記動作候補を入力とし、前記ロボットが第1の処理動作を行うときの、前記各動作候補の評価を示す第1の評価値を出力するために、第1学習データにより学習された学習済みの第1の学習器と、
前記撮影画像、及び少なくとも一つの前記動作候補を入力とし、前記ロボットが第2の処理動作を行うときの、前記各動作候補の評価を示す第2の評価値を出力するために、前記第1学習データとは異なる第2学習データにより、学習された学習済みの第2の学習器と、
前記第1の評価値及び前記第2の評価値の少なくとも一方に基づいて、前記ロボットを動作させるための指令値を算出する評価部と、
を備えている、動作制御装置。 - 前記第1の評価値は、当該第1の動作候補の成功確率を示す確率値であり、
前記第2の評価値は、当該第2の動作候補の成功確率を示す確率値である、請求項1に記載の動作制御装置。 - 前記評価部は、前記各評価値が所定値以上の前記動作候補に基づいて、前記指令値を算出する、請求項1または2に記載の動作制御装置。
- 前記評価部は、前記第1の評価値及び第2の評価値に対し、前記各処理動作に基づく重み付けを行うことで、前記指令値を算出する、請求項1または2に記載の動作制御装置。
- 前記第2の評価値は、前記第1の処理動作及び第2の処理動作の少なくとも一方に起因して起こり得る事象の評価値である、請求項1から4のいずれかに記載の動作制御装置。
- 前記撮影画像、及び少なくとも一つの前記動作候補を入力とし、前記第1の処理動作及び第2の処理動作の少なくとも一方に起因して起こり得る事象を示す第3の評価値を出力とするように、前記第1及び第2学習データとは異なる第3学習データにより、学習された学習済みの第3の学習器をさらに備え、
前記評価部は、前記第3の評価値も用いて、前記指令値を算出する、請求項1から4のいずれかに記載の動作制御装置。 - 前記撮影画像を入力とし、前記第1の処理動作に適した前記第1の動作候補、及び前記第2の処理動作に適した前記第2の動作候補を出力とする動作候補生成部をさらに備えている、請求項1から6のいずれかに記載の動作制御装置。
- 前記第1の処理動作及び前記第2の処理動作の動作候補は、前記ロボットが動作するための移動ベクトルを含む、請求項1から7のいずれかに記載の動作制御装置。
- 処理対象物に対して、処理を行うロボットと、
請求項1から8のいずれかに記載の動作制御装置と、
を備え、
前記動作制御装置から算出された指令値により、前記ロボットが動作するように構成されている、ロボット制御システム。 - 処理対象物に対するロボットの動作制御方法であって、
少なくとも前記処理対象物を撮影した撮影画像、及び前記ロボットの少なくとも一つの動作候補を入力するステップと、
前記撮影画像、及び少なくとも一つの前記動作候補を入力とし、前記ロボットが第1の処理動作を行うときの、前記各動作候補の評価を示す第1の評価値を出力するために、第1学習データにより、学習された学習済みの第1の学習器を準備するステップと、
前記撮影画像、及び少なくとも一つの前記動作候補を入力とし、前記ロボットが第2の処理動作を行うときの、前記各動作候補の評価を示す第2の評価値を出力するために、前記第1学習データとは異なる第2学習データにより、学習された学習済みの第2の学習器を準備するステップと、
前記撮影画像及び前記動作候補から、前記第1の学習器によって、前記第1の評価値を出力するステップと、
前記撮影画像及び前記動作候補から、前記第2の学習器によって、前記第2の評価値を出力するステップと、
前記第1の評価値及び第2の評価値の少なくとも一方に基づいて、前記ロボットを動作させるための指令値を算出するステップと、
を備えている、動作制御方法。 - 処理対象物に対するロボットの動作制御プログラムであって、
コンピュータに、
少なくとも前記処理対象物を撮影した撮影画像、及び前記ロボットの少なくとも一つの動作候補を入力するステップと、
前記撮影画像、及び少なくとも一つの前記動作候補を入力とし、前記ロボットが第1の処理動作を行うときの、前記各動作候補の評価を示す第1の評価値を出力するために、第1学習データにより、学習された学習済みの第1の学習器を準備するステップと、
前記撮影画像、及び少なくとも一つの前記動作候補を入力とし、前記ロボットが第2の処理動作を行うときの、前記各動作候補の評価を示す第2の評価値を出力するために、前記第1学習データとは異なる第2学習データにより、学習された学習済みの第2の学習器を準備するステップと、
前記撮影画像及び前記動作候補から、前記第1の学習器によって、前記第1の評価値を出力するステップと、
前記撮影画像及び前記動作候補から、前記第2の学習器によって、前記第2の評価値を出力するステップと、
前記第1の評価値及び第2の評価値の少なくとも一方に基づいて、前記ロボットを動作させるための指令値を算出するステップと、
を実行させる、動作制御プログラム。 - 制御対象物の制御装置であって、
前記制御対象物に関連する環境情報、及び前記制御対象物に対する少なくとも一つの処理候補を入力する入力部と、
前記環境情報、及び少なくとも一つの前記処理候補を入力とし、前記制御対象物に対し第1の処理を行うときの、前記各処理候補の評価を示す第1の評価値を出力するために、第1学習データにより、学習された学習済みの第1の学習器と、
前記環境情報、及び少なくとも一つの前記処理候補を入力とし、前記制御対象物に対し第2の処理を行うときの、前記各処理候補の評価を示す第2の評価値を出力するために、前記第1学習データとは異なる第2学習データにより、学習された学習済みの第2の学習器と、
前記第1の評価値及び前記第2の評価値の少なくとも一方に基づいて、前記制御対象物を制御するための指令値を算出する評価部と、
を備えている、制御装置。 - 所定の処理を行う処理装置であって、
少なくとも一つの処理候補を入力する入力部と、
少なくとも一つの処理候補を入力とし、第1の処理を行うときの、前記各処理候補の評価を示す第1の評価値を出力するために、第1学習データにより、学習された学習済みの第1の学習器と、
少なくとも一つの前記処理候補を入力とし、第2の処理を行うときの、前記各処理候補の評価を示す第2の評価値を出力するために、前記第1学習データとは異なる第2学習データにより、学習された学習済みの第2の学習器と、
前記第1の評価値及び前記第2の評価値の少なくとも一方に基づいて、前記第1の処理及び前記第2の処理の少なくとも一方の処理候補を決定する評価部と、
を備えている、処理装置。
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| EP19766953.4A EP3766643A4 (en) | 2018-03-15 | 2019-02-20 | OPERATION CONTROL DEVICE FOR ROBOTS |
| CN201980012028.8A CN111683799B (zh) | 2018-03-15 | 2019-02-20 | 动作控制装置、系统、方法、存储介质、控制及处理装置 |
| US16/977,478 US11478926B2 (en) | 2018-03-15 | 2019-02-20 | Operation control device for robot, robot control system, operation control method, control device, processing device and recording medium |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2018-047704 | 2018-03-15 | ||
| JP2018047704A JP6911798B2 (ja) | 2018-03-15 | 2018-03-15 | ロボットの動作制御装置 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2019176478A1 true WO2019176478A1 (ja) | 2019-09-19 |
Family
ID=67907709
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/JP2019/006183 Ceased WO2019176478A1 (ja) | 2018-03-15 | 2019-02-20 | ロボットの動作制御装置 |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US11478926B2 (ja) |
| EP (1) | EP3766643A4 (ja) |
| JP (1) | JP6911798B2 (ja) |
| CN (1) | CN111683799B (ja) |
| WO (1) | WO2019176478A1 (ja) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP4260994A4 (en) * | 2020-12-08 | 2024-10-09 | Sony Group Corporation | LEARNING DEVICE, LEARNING SYSTEM, AND LEARNING METHOD |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP7204513B2 (ja) * | 2019-02-13 | 2023-01-16 | 株式会社東芝 | 制御装置及びプログラム |
| JP7378309B2 (ja) * | 2020-02-10 | 2023-11-13 | 住友重機械工業株式会社 | 作業装置 |
| JP7234970B2 (ja) * | 2020-02-17 | 2023-03-08 | 株式会社デンソー | 車両行動生成装置、車両行動生成方法、および車両行動生成プログラム |
| JP7347252B2 (ja) | 2020-02-17 | 2023-09-20 | 株式会社デンソー | 車両行動評価装置、車両行動評価方法、および車両行動評価プログラム |
| DE102020212658A1 (de) * | 2020-10-07 | 2022-04-07 | Robert Bosch Gesellschaft mit beschränkter Haftung | Vorrichtung und Verfahren zum Steuern einer Robotervorrichtung |
| TWI888681B (zh) * | 2020-12-25 | 2025-07-01 | 日商發那科股份有限公司 | 機械學習裝置、及機械學習方法 |
| WO2023042307A1 (ja) * | 2021-09-15 | 2023-03-23 | ヤマハ発動機株式会社 | 画像処理装置、部品把持システム、画像処理方法および部品把持方法 |
| CN113829358B (zh) * | 2021-11-01 | 2022-12-27 | 江苏昱博自动化设备有限公司 | 一种基于深度强化学习的机器人抓取多目标物的训练方法 |
| JP2023180680A (ja) * | 2022-06-10 | 2023-12-21 | セイコーエプソン株式会社 | 軌道作成方法および軌道作成装置 |
| CN116810780A (zh) * | 2023-06-13 | 2023-09-29 | 中国南方电网有限责任公司超高压输电公司电力科研院 | 巡检机器人的巡检方式确定方法、装置、设备和存储介质 |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH05101028A (ja) * | 1991-10-04 | 1993-04-23 | Nippon Telegr & Teleph Corp <Ntt> | 複数特徴量の統合判定方法 |
| JP2014240110A (ja) * | 2013-06-12 | 2014-12-25 | 三菱電機株式会社 | ビンピッキング性能評価装置及び方法 |
| JP2017030135A (ja) * | 2015-07-31 | 2017-02-09 | ファナック株式会社 | ワークの取り出し動作を学習する機械学習装置、ロボットシステムおよび機械学習方法 |
| JP2017097317A (ja) * | 2015-11-18 | 2017-06-01 | 本田技研工業株式会社 | 識別装置、ロボットおよび識別方法 |
| JP2017185578A (ja) | 2016-04-05 | 2017-10-12 | 株式会社リコー | 物体把持装置及び把持制御プログラム |
| JP2017185577A (ja) * | 2016-04-04 | 2017-10-12 | ファナック株式会社 | シミュレーション結果を利用して学習を行う機械学習装置,機械システム,製造システムおよび機械学習方法 |
| WO2018008503A1 (ja) * | 2016-07-05 | 2018-01-11 | 株式会社日立製作所 | 制御装置 |
| JP2018027581A (ja) * | 2016-08-17 | 2018-02-22 | 株式会社安川電機 | ピッキングシステム |
Family Cites Families (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006139185A (ja) * | 2004-11-15 | 2006-06-01 | Nippon Telegr & Teleph Corp <Ntt> | 音声認識方法、この方法を実施する装置、プログラムおよびその記録媒体 |
| FI20105732A0 (fi) | 2010-06-24 | 2010-06-24 | Zenrobotics Oy | Menetelmä fyysisten kappaleiden valitsemiseksi robottijärjestelmässä |
| JP5754306B2 (ja) * | 2011-09-01 | 2015-07-29 | 富士ゼロックス株式会社 | 画像識別情報付与プログラム及び画像識別情報付与装置 |
| US10518409B2 (en) * | 2014-09-02 | 2019-12-31 | Mark Oleynik | Robotic manipulation methods and systems for executing a domain-specific application in an instrumented environment with electronic minimanipulation libraries |
| US9630318B2 (en) | 2014-10-02 | 2017-04-25 | Brain Corporation | Feature detection apparatus and methods for training of robotic navigation |
| US11049045B2 (en) * | 2015-11-18 | 2021-06-29 | Honda Motor Co., Ltd. | Classification apparatus, robot, and classification method |
| CA3016418C (en) | 2016-03-03 | 2020-04-14 | Google Llc | Deep machine learning methods and apparatus for robotic grasping |
| EP3550498A4 (en) * | 2016-09-07 | 2020-05-27 | Bosch Corporation | INFORMATION PROCESSING DEVICE AND INFORMATION PROCESSING SYSTEM |
| DE202017106132U1 (de) * | 2016-10-10 | 2017-11-13 | Google Llc | Neuronale Netze zum Auswählen von durch einen Roboteragenten auszuführenden Aktionen |
| JP6453922B2 (ja) * | 2017-02-06 | 2019-01-16 | ファナック株式会社 | ワークの取り出し動作を改善するワーク取り出し装置およびワーク取り出し方法 |
| JP6951659B2 (ja) * | 2017-05-09 | 2021-10-20 | オムロン株式会社 | タスク実行システム、タスク実行方法、並びにその学習装置及び学習方法 |
| JP6546618B2 (ja) * | 2017-05-31 | 2019-07-17 | 株式会社Preferred Networks | 学習装置、学習方法、学習モデル、検出装置及び把持システム |
| CN107688856B (zh) * | 2017-07-24 | 2020-11-06 | 清华大学 | 基于深度强化学习的室内机器人场景主动识别方法 |
| JP6579498B2 (ja) * | 2017-10-20 | 2019-09-25 | 株式会社安川電機 | 自動化装置及び位置検出装置 |
-
2018
- 2018-03-15 JP JP2018047704A patent/JP6911798B2/ja active Active
-
2019
- 2019-02-20 US US16/977,478 patent/US11478926B2/en active Active
- 2019-02-20 EP EP19766953.4A patent/EP3766643A4/en active Pending
- 2019-02-20 CN CN201980012028.8A patent/CN111683799B/zh active Active
- 2019-02-20 WO PCT/JP2019/006183 patent/WO2019176478A1/ja not_active Ceased
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH05101028A (ja) * | 1991-10-04 | 1993-04-23 | Nippon Telegr & Teleph Corp <Ntt> | 複数特徴量の統合判定方法 |
| JP2014240110A (ja) * | 2013-06-12 | 2014-12-25 | 三菱電機株式会社 | ビンピッキング性能評価装置及び方法 |
| JP2017030135A (ja) * | 2015-07-31 | 2017-02-09 | ファナック株式会社 | ワークの取り出し動作を学習する機械学習装置、ロボットシステムおよび機械学習方法 |
| JP2017097317A (ja) * | 2015-11-18 | 2017-06-01 | 本田技研工業株式会社 | 識別装置、ロボットおよび識別方法 |
| JP2017185577A (ja) * | 2016-04-04 | 2017-10-12 | ファナック株式会社 | シミュレーション結果を利用して学習を行う機械学習装置,機械システム,製造システムおよび機械学習方法 |
| JP2017185578A (ja) | 2016-04-05 | 2017-10-12 | 株式会社リコー | 物体把持装置及び把持制御プログラム |
| WO2018008503A1 (ja) * | 2016-07-05 | 2018-01-11 | 株式会社日立製作所 | 制御装置 |
| JP2018027581A (ja) * | 2016-08-17 | 2018-02-22 | 株式会社安川電機 | ピッキングシステム |
Non-Patent Citations (1)
| Title |
|---|
| See also references of EP3766643A4 |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP4260994A4 (en) * | 2020-12-08 | 2024-10-09 | Sony Group Corporation | LEARNING DEVICE, LEARNING SYSTEM, AND LEARNING METHOD |
Also Published As
| Publication number | Publication date |
|---|---|
| US20210053214A1 (en) | 2021-02-25 |
| JP6911798B2 (ja) | 2021-07-28 |
| CN111683799A (zh) | 2020-09-18 |
| EP3766643A4 (en) | 2022-01-12 |
| US11478926B2 (en) | 2022-10-25 |
| CN111683799B (zh) | 2024-01-12 |
| JP2019155561A (ja) | 2019-09-19 |
| EP3766643A1 (en) | 2021-01-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6911798B2 (ja) | ロボットの動作制御装置 | |
| CN108873768B (zh) | 任务执行系统及方法、学习装置及方法、以及记录介质 | |
| JP2025517823A (ja) | ロボットの自然言語制御 | |
| JP7517225B2 (ja) | 軌道生成システム、軌道生成方法及びプログラム | |
| CN114080304A (zh) | 控制装置、控制方法及控制程序 | |
| US20250178615A1 (en) | Robot navigation using a high-level policy model and a trained low-level policy model | |
| CN115351780A (zh) | 用于控制机器人设备的方法 | |
| WO2020110505A1 (ja) | 画像生成装置、ロボット訓練システム、画像生成方法、及び画像生成プログラム | |
| JP6933167B2 (ja) | ロボットの制御装置 | |
| CN117295589A (zh) | 在训练和细化机器人控制策略中使用模仿学习的系统和方法 | |
| CN114387513A (zh) | 机器人抓取方法、装置、电子设备及存储介质 | |
| CN116587275A (zh) | 基于深度强化学习的机械臂智能阻抗控制方法及系统 | |
| Baxter et al. | Deep prediction of swept volume geometries: Robots and resolutions | |
| Zheng et al. | Deep visual-guided and deep reinforcement learning algorithm based for multip-peg-in-hole assembly task of power distribution live-line operation robot | |
| JP7468619B2 (ja) | 学習装置、学習方法、及び、記録媒体 | |
| Lee et al. | UaMPNet: Uncertainty-Aware Motion Planning Network for Manipulator Motion Planning | |
| US20230316734A1 (en) | Pose fusion estimation | |
| Tran et al. | Predicting sample collision with neural networks | |
| US20250037025A1 (en) | Method for training a machine learning model for controlling a robot to manipulate an object | |
| JP2025019043A (ja) | センサデータを分類するための機械学習モデルをトレーニングするための方法 | |
| CN118478373A (zh) | 用于控制机器人装置的装置和方法 | |
| Hong et al. | Dynamics-aware metric embedding: Metric learning in a latent space for visual planning | |
| CN120461440B (zh) | 面向复杂操作任务的双臂机器人运动规划方法和系统 | |
| US20250328397A1 (en) | Spatio-temporal graph and message passing | |
| CN120039279B (zh) | 面向人形机器人的多风格车辆自动驾驶方法 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19766953 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2019766953 Country of ref document: EP |
|
| ENP | Entry into the national phase |
Ref document number: 2019766953 Country of ref document: EP Effective date: 20201015 |
|
| WWW | Wipo information: withdrawn in national office |
Ref document number: 2019766953 Country of ref document: EP |