CN106886165A - A kind of manipulator crawl and method of operating based on simulator - Google Patents
A kind of manipulator crawl and method of operating based on simulator Download PDFInfo
- Publication number
- CN106886165A CN106886165A CN201710141949.7A CN201710141949A CN106886165A CN 106886165 A CN106886165 A CN 106886165A CN 201710141949 A CN201710141949 A CN 201710141949A CN 106886165 A CN106886165 A CN 106886165A
- Authority
- CN
- China
- Prior art keywords
- crawl
- candidate
- image
- manipulator
- simulator
- 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.)
- Pending
Links
Classifications
-
- 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
- G05B17/00—Systems involving the use of models or simulators of said systems
- G05B17/02—Systems involving the use of models or simulators of said systems electric
Abstract
A kind of the manipulator crawl and method of operating based on simulator proposed in the present invention, its main contents include：Simulation initialization, execution artificial tasks and data acquisition, later stage treatment, its process is, first pass through initial object, the configuration of hand, attribute and crawl candidate data storehouse and be simulated initialization, then mapped by the crawl of different images and data set is created in feasible time quantum by some form of parallelization, perform artificial tasks and data acquisition, aggregated data and the pretreatment of object are finally based on, object crawl database is formed, and carry out later stage treatment.Manipulator crawl and method of operating of the present invention based on simulator, reduce noise, are limited by sensor and also greatly reduced；Flexibility, the stability of manipulator are improve, the accuracy of measurement is also improved.
Description
Technical field
The present invention relates to field in intelligent robotics, more particularly, to a kind of manipulator crawl and manipulation based on simulator
Method.
Background technology
Robot delicate can complete flexible, fine grasping manipulation, more as effective extension of mankind's activity limbs
More to turn into one of the popular research direction of robot field.Relative to simple end-effector, robot delicate has
Highly versatile, perception are abundant, can realize meeting geometry closing and the advantages of force-closed accurate, firm crawl.It is
One it is highly integrated, with various perceptional functions and intelligentized Mechatronic Systems, be related to theory of mechanisms, bionics, automatically control,
The multiple research field such as sensor technology, computer technology, artificial intelligence, mechanics of communication, microelectronics, materialogy and intersection are learned
Section.
Manipulator can be widely applied in hazardous environment operation, marine resources detection and space exploration, in the future also will be by
In gradually spreading to our daily life.But current manipulator is due to being limited (including noise) by sensor, generally, survey
Amount result is all not accurate enough.
The present invention proposes a kind of manipulator crawl based on simulator and method of operating, first passes through initial object, hand
Configuration, attribute and crawl candidate data storehouse are simulated initialization, are then mapped and by certain by the crawl of different images
The parallelization of form creates data set in feasible time quantum, performs artificial tasks and data acquisition, is finally based on object
Aggregated data and pretreatment, form object crawl database, and carry out later stage treatment.The present invention reduces noise, sensed
The limitation of device is also greatly reduced；Flexibility, the stability of manipulator are improve, the accuracy of measurement is also improved.
The content of the invention
For the inaccurate problem of measurement result, grabbed it is an object of the invention to provide a kind of manipulator based on simulator
Take and method of operating, first pass through initial object, the configuration of hand, attribute and crawl candidate data storehouse and be simulated initialization, then
By the crawl mapping and parallelization of different images, artificial tasks and data acquisition are performed, be finally based on the aggregated data of object
And pretreatment, form object crawl database.
To solve the above problems, the present invention provides a kind of manipulator crawl based on simulator and method of operating, and its is main
Content includes：
(1) simulation initialization；
(2) artificial tasks and data acquisition are performed；
(3) later stage treatment.
Wherein, described simulation initialization, each simulation needs the configuration of an initial object and hand；Object properties need
It is defined, and needs to generate possible crawl candidate list；Including initial object, the configuration of hand, attribute and crawl candidate's number
According to storehouse.
Further, the configuration of described initial object and hand and attribute, since all object grids are pre-processed；Each
Object grid is loaded into Python scripts, and obtains the estimation in object quality and the center of inertia；
Using these pretreatment values, each grid is loaded into robot simulation simulation softward (V-REP), determines object
Rest attitude and handgrip initial attitude；First for object distributes a bounding box, the bounding box passes through relative { W } weight
The posture of new orientation object, is assigned as the geometric center of object to estimate the posture of object by frame center；Then along { T } just
Z-direction placing objects 0.3m；Relative to { T }, object is set to be centered in (x, y)=(0.0), remains stationary appearance using pure translational component
State；
The stationary posture is given, the initial position being placed on along positive Z-direction in { O } then will be captured；Chosen distance pair
The center of elephantRice, from Partial frame to along x, the bounding box edge direction of y, z；Record all objects
Attribute (including subject poses, object bounds frame and material) and crawl attitude, and this mistake is repeated to each object that data are concentrated
Journey.
Further, described crawl candidate data storehouse, in simulations for covering the possible side for capturing candidate spatial
Method, based on the front and rear multiplication of object configuration, it is represented as transformation matrix；
The bounding box and fixture postures of given object, by around object globally rotary grasping (pre-multiplied) and locally (after
Multiply) calculate offline crawl candidate；Respectively in X, 3 × 3 spin matrix R are multiplied by Y and Z axisX(α)RY(β)RZ(γ)；Omit α,
Beta, gamma, transformation matrix is calculated according to following formula：
Wherein, Q represents the final conversion of crawl coordinate system；Crawl candidate item off-line execution in Python scripts is calculated,
And use the estimation bounding box of object, transformation matrixWithSelection constraint makes around Z axis 8 rotations of (i.e. every 45 °) generation,
And locally rotating will occur with the yardstick somewhat thinner than overall situation rotation.
Further, described crawl candidate, after computing formula (1), check whether new chucking position suitable (if
Be, cancel clamp candidate), solve system of linear equations, check the vectorial normal from fixture palm whether the bounding box with object
It is intersecting；If intersecting, crawl candidate is added in crawl candidate data storehouse, and repeats the process, until rotation list is used up；
In all possible candidate item in database, most 10000 checkings are selected in simulator.
Wherein, described execution artificial tasks and data acquisition, including different images crawl mapping and parallelization；Pass through
Object is loaded into simulation, and is initialized its quality, inertia and attitude in the value that initial phase is recorded and is started；Object
It is initially placed into static step so that when finger tip and object contact, object is not moved；After loading object, simulator exists
Initial phase is sampled to test to potential candidate item subset；It is due to crawl configuration and latent with workbench or object
In collision, if crawl is infeasible, stopping currently being attempted and moved at next candidate.
Further, described test, using proximity transducer check each feasible crawl candidate verify palm towards
Object；If in the position, connecting the proximity transducer detection object of crawl, then the detected surface point of its record, and away from
It is with a distance from the surface point for detecting：Three crawls (capturing orientation using identical) are attempted at 0.06,0.09,0.12m, and
Along original palm normal；These distances are selected as the finger tip in palm and Barrett Hand (Barrett manipulator)
The distance between (0.145m) is interior, and allows to detect the geometry of object with slightly different yardstick；
During attempting each time, camera is positioned at the distance of palm 0.25m along local negative Z-direction, and
And the image of object was recorded before crawl is attempted；Once crawl is placed and have recorded image, executor is around object
Closure；If all finger tips are contacted with object, object is changed into dynamic analog, and since elevation process；
The target raised position relative to { T } (0.0,0.0,0.60m) is selected, and forces manipulator to be protected during advancing
Hold current grip posture；Once crawl has arrived at target location, if all finger tips are still contacted with object, then it is assumed that hold
Hold is stabilization and success；The process is repeated, until crawl candidate list is used up；In V-REP, motor pool wrapper is used
To calculate track, and along the path execution incremental steps of generation.
Further, described different image and crawl map, due to crawl programming always in a similar way around
Object is closed, so two different views of object are collected in clamping process being respectively：
(1) direction of camera is always upward (one-to-many mapping)；
(2) direction of camera always matches the direction (one-to-one mapping) of crawl；
Be incorporated into ambiguity in grasping space by the one-to-many mapping between by arousing image and grasping；This
In the case of, fixture orientation is not directly related to camera orientation, it means that single image can correspond to may many differences
Fixture；However, more direct relation between introducing image and grasping；The similar orientation reflection of the object for capturing in the picture
Similar orientation in grasping.
Further, described parallelization, because this large amount of crawl candidate is sampled, and the object to be assessed
Quantity is relatively high, in order to create data set in feasible time quantum, it is necessary to some form of parallelization；
Due to the requirement from the vision sensor for needing a small amount of memory from graphics card, under server, not have
The mode operation for having any graphical interfaces operates each scene.
Wherein, described later stage treatment, in simulations, the information captured by depth buffer be encoded into [0,1] it
Between scope, and can by it is following operation be decoded as actual value：
I=Xnear+I*(Xfar-Xnear) (2)
Wherein, I is the image collected, Xnear, XfarIt is respectively that near and far cuts out plane；Delete all object crawl examples
Right, wherein image variance is less than 1e-3；When camera heights and table height match, all crawls of collected image are deleted.
Brief description of the drawings
Fig. 1 is a kind of system flow chart of manipulator crawl and method of operating based on simulator of the present invention.
Fig. 2 is a kind of crawl candidate data storehouse of manipulator crawl and method of operating based on simulator of the present invention.
Fig. 3 is a kind of flow of the execution artificial tasks of manipulator crawl and method of operating based on simulator of the present invention
Figure.
Fig. 4 is a kind of data acquisition of manipulator crawl and method of operating based on simulator of the present invention.
Specific embodiment
It should be noted that in the case where not conflicting, the feature in embodiment and embodiment in the application can phase
Mutually combine, the present invention is described in further detail with specific embodiment below in conjunction with the accompanying drawings.
Fig. 1 is a kind of system flow chart of manipulator crawl and method of operating based on simulator of the present invention.Mainly include
Simulation initialization, performs artificial tasks and data acquisition, later stage treatment.
Simulation initialization, each simulation needs the configuration of an initial object and hand；Object properties need to be defined, and
Need to generate possible crawl candidate list；Including initial object, the configuration of hand, attribute and crawl candidate data storehouse.
Wherein, the configuration of initial object and hand and attribute, since all object grids are pre-processed；Each object grid quilt
It is loaded into Python scripts, and obtains the estimation in object quality and the center of inertia；
Using these pretreatment values, each grid is loaded into robot simulation simulation softward (V-REP), determines object
Rest attitude and handgrip initial attitude；First for object distributes a bounding box, the bounding box passes through relative { W } weight
The posture of new orientation object, is assigned as the geometric center of object to estimate the posture of object by frame center；Then along { T } just
Z-direction placing objects 0.3m；Relative to { T }, object is set to be centered in (x, y)=(0.0), remains stationary appearance using pure translational component
State；
The stationary posture is given, the initial position being placed on along positive Z-direction in { O } then will be captured；Chosen distance pair
The center of elephantRice, from Partial frame to along x, the bounding box edge direction of y, z；Record all objects
Attribute (including subject poses, object bounds frame and material) and crawl attitude, and this mistake is repeated to each object that data are concentrated
Journey.
Later stage is processed, and in simulations, the information captured by depth buffer is encoded into the scope between [0,1], and
And actual value can be decoded as by following operation：
I=Xnear+I*(Xfar-Xnear) (2)
Wherein, I is the image collected, Xnear, XfarIt is respectively that near and far cuts out plane；Delete all object crawl examples
Right, wherein image variance is less than 1e-3；When camera heights and table height match, all crawls of collected image are deleted.
Fig. 2 is a kind of crawl candidate data storehouse of manipulator crawl and method of operating based on simulator of the present invention.In mould
The method for being used to cover possible crawl candidate spatial in plan, based on the front and rear multiplication of object configuration, it is represented as converting square
Battle array；
The bounding box and fixture postures of given object, by around object globally rotary grasping (pre-multiplied) and locally (after
Multiply) calculate offline crawl candidate；Respectively in X, 3 × 3 spin matrix R are multiplied by Y and Z axisX(α)RY(β)RZ(γ)；Omit α,
Beta, gamma, transformation matrix is calculated according to following formula：
Wherein, Q represents the final conversion of crawl coordinate system；Crawl candidate item off-line execution in Python scripts is calculated,
And use the estimation bounding box of object, transformation matrixWithSelection constraint makes around Z axis 8 rotations of (i.e. every 45 °) generation,
And locally rotating will occur with the yardstick somewhat thinner than overall situation rotation.
After computing formula (1), check whether new chucking position is suitable (if it is, cancel clamping candidate), solves line
Property equation group, check the vectorial normal from fixture palm whether intersect with the bounding box of object；If intersecting, candidate will be captured
It is added in crawl candidate data storehouse, and repeats the process, until rotation list is used up；All possible time in database
In option, most 10000 checkings are selected in simulator.
Fig. 3 is a kind of flow of the execution artificial tasks of manipulator crawl and method of operating based on simulator of the present invention
Figure.Performing artificial tasks and data acquisition includes the crawl mapping and parallelization of different images；Simulation is loaded into by by object
In, and initialize its quality, inertia and attitude in the value that initial phase is recorded and start；Object is initially placed into static state
In step so that when finger tip and object contact, object is not moved；After loading object, simulator is in initial phase to potential
Candidate item subset is sampled to be tested；Potential collision due to crawl configuration and with workbench or object, if crawl is not
It is feasible, then stop currently attempting and moving at next candidate.
Fig. 4 is a kind of data acquisition of manipulator crawl and method of operating based on simulator of the present invention.Using connecing
Nearly sensor checks that each feasible crawl candidate verifies palm towards object；If in the position, connection crawl close to biography
Sensor detection object, then it records detected surface point, and the distance of the surface point detected in distance is：0.06,0.09,
Three crawls (capturing orientation using identical) are attempted at 0.12m, and along original palm normal；These distances are selected as
The distance between finger tip (0.145m) in palm and Barrett Hand (Barrett manipulator) is interior, and allows with slightly different
Yardstick detect the geometry of object；
During attempting each time, camera is positioned at the distance of palm 0.25m along local negative Z-direction, and
And the image of object was recorded before crawl is attempted；Once crawl is placed and have recorded image, executor is around object
Closure；If all finger tips are contacted with object, object is changed into dynamic analog, and since elevation process；
The target raised position relative to { T } (0.0,0.0,0.60m) is selected, and forces manipulator to be protected during advancing
Hold current grip posture；Once crawl has arrived at target location, if all finger tips are still contacted with object, then it is assumed that hold
Hold is stabilization and success；The process is repeated, until crawl candidate list is used up；In V-REP, motor pool wrapper is used
To calculate track, and along the path execution incremental steps of generation.
Because crawl programming is always closed around object in a similar way, so collecting the two of object in clamping process
Individual different views are respectively：
(1) direction of camera is always upward (one-to-many mapping)；
(2) direction of camera always matches the direction (one-to-one mapping) of crawl；
Be incorporated into ambiguity in grasping space by the one-to-many mapping between by arousing image and grasping；This
In the case of, fixture orientation is not directly related to camera orientation, it means that single image can correspond to may many differences
Fixture；However, more direct relation between introducing image and grasping；The similar orientation reflection of the object for capturing in the picture
Similar orientation in grasping.
Because this large amount of crawl candidate is sampled, and the quantity of the object to be assessed is relatively high, in order to feasible
Time quantum in create data set, it is necessary to some form of parallelization；
Due to the requirement from the vision sensor for needing a small amount of memory from graphics card, under server, not have
The mode operation for having any graphical interfaces operates each scene.
For those skilled in the art, the present invention is not restricted to the details of above-described embodiment, without departing substantially from essence of the invention
In the case of god and scope, the present invention can be realized with other concrete forms.Additionally, those skilled in the art can be to this hair
Bright to carry out various changes and modification without departing from the spirit and scope of the present invention, these improvement also should be regarded as of the invention with modification
Protection domain.Therefore, appended claims are intended to be construed to include preferred embodiment and fall into all changes of the scope of the invention
More and modification.
Claims (10)
1. a kind of manipulator crawl and method of operating based on simulator, it is characterised in that main to include simulation initialization ()；
Perform artificial tasks and data acquisition (two)；Later stage processes (three).
2. based on simulation initialization () described in claims 1, it is characterised in that each simulation needs an initial object
With the configuration of hand；Object properties need to be defined, and need to generate possible crawl candidate list；Including initial object, hand
Configuration, attribute and crawl candidate data storehouse.
3. based on the initial object and configuration and the attribute of hand described in claims 2, it is characterised in that all right from pre-processing
As grid starts；Each object grid is loaded into Python scripts, and obtains the estimation in object quality and the center of inertia；
Using these pretreatment values, each grid is loaded into robot simulation simulation softward (V-REP), determines the first of object
The initial attitude of beginning stationary posture and handgrip；First for object distributes a bounding box, it is again fixed that the bounding box passes through relative { W }
To the posture of object, frame center is assigned as into the geometric center of object to estimate the posture of object；Then along the positive Z side of { T }
To placing objects 0.3m；Relative to { T }, object is set to be centered in (x, y)=(0.0), remains stationary attitude using pure translational component；
The stationary posture is given, the initial position being placed on along positive Z-direction in { O } then will be captured；Chosen distance object
CenterRice, from Partial frame to along x, the bounding box edge direction of y, z；Record all object properties
(including subject poses, object bounds frame and material) and crawl attitude, and this process is repeated to each object that data are concentrated.
4. based on the crawl candidate data storehouse described in claims 2, it is characterised in that in simulations for covering possible grabbing
The method for taking candidate spatial, based on the front and rear multiplication of object configuration, it is represented as transformation matrix；
The bounding box and fixture postures of given object, by around object globally rotary grasping (pre-multiplied) and locally (multiplying afterwards) come
Calculate offline crawl candidate；Respectively in X, 3 × 3 spin matrix R are multiplied by Y and Z axisX(α)RY(β)RZ(γ)；Omission α, beta, gamma,
Transformation matrix is calculated according to following formula：
Wherein, Q represents the final conversion of crawl coordinate system；Crawl candidate item off-line execution in Python scripts is calculated, and is made
With the estimation bounding box of object, transformation matrixWithSelection constraint makes around Z axis 8 rotations of (i.e. every 45 °) generation, and
Local rotation will occur with the yardstick somewhat thinner than overall situation rotation.
5., based on the crawl candidate described in claims 4, it is characterised in that after computing formula (1), new fixture position is checked
Put whether properly (if it is, cancel clamping candidate), solve system of linear equations, whether check the vectorial normal from fixture palm
Intersect with the bounding box of object；If intersecting, crawl candidate is added in crawl candidate data storehouse, and repeats the process, directly
Used up to rotation list；In all possible candidate item in database, most 10000 checkings are selected in simulator.
6. based on the execution artificial tasks described in claims 1 and data acquisition (two), it is characterised in that including different images
Crawl mapping and parallelization；It is loaded into simulation by by object, and its matter is initialized in the value that initial phase is recorded
Amount, inertia and attitude and start；Object is initially placed into static step so that when finger tip and object contact, object is not
It is mobile；After loading object, simulator is sampled to test in initial phase to potential candidate item subset；Due to crawl
Configuration and the potential collision with workbench or object, if crawl is infeasible, stopping currently being attempted and moves to next candidate
Place.
7. based on the test described in claims 6, it is characterised in that check that each feasible crawl is waited using proximity transducer
Palm towards object is verified in choosing；If in the position, connecting the proximity transducer detection object of crawl, then its record is detected
Surface point, and the distance of the surface point detected in distance is：Three crawls are attempted at 0.06,0.09,0.12m (using identical
Crawl orientation), and along original palm normal；These distances are selected as in palm and Barrett Hand (Barretts
Manipulator) the distance between finger tip (0.145m) it is interior, and allow to detect the geometry of object with slightly different yardstick；
During attempting each time, camera is positioned at the distance of palm 0.25m along local negative Z-direction, and
The image of object is recorded before attempting crawl；Once crawl is placed and have recorded image, executor is closed around object；
If all finger tips are contacted with object, object is changed into dynamic analog, and since elevation process；
The target raised position relative to { T } (0.0,0.0,0.60m) is selected, and forces manipulator to keep working as during advancing
Preceding grip posture；Once crawl has arrived at target location, if all finger tips are still contacted with object, then it is assumed that gripping is
It is stabilization and successful；The process is repeated, until crawl candidate list is used up；In V-REP, counted using motor pool wrapper
Track is calculated, and along the path execution incremental steps of generation.
8. mapped based on the different image described in claims 6 and crawl, it is characterised in that due to crawl programming always with
Similar mode is closed around object, so two different views of object are collected in clamping process being respectively：
(1) direction of camera is always upward (one-to-many mapping)；
(2) direction of camera always matches the direction (one-to-one mapping) of crawl；
Be incorporated into ambiguity in grasping space by the one-to-many mapping between by arousing image and grasping；In such case
Under, fixture orientation is not directly related to camera orientation, it means that single image can correspond to may many different folders
Tool；However, more direct relation between introducing image and grasping；The similar orientation of the object for capturing in the picture is reflected grabs
Hold interior similar orientation.
9. based on the parallelization described in claims 6, it is characterised in that because this large amount of crawl candidate is sampled, and
And the quantity of the object to be assessed is relatively high, in order to create data set in feasible time quantum, it is necessary to some form of parallel
Change；
Due to the requirement from the vision sensor for needing a small amount of memory from graphics card, under server, appointed with no
The mode operation of what graphical interfaces operates each scene.
10. based on later stage treatment (three) described in claims 1, it is characterised in that in simulations, caught by depth buffer
The information for obtaining is encoded into the scope between [0,1], and can be decoded as actual value by following operation：
I=Xnear+I*(Xfar-Xnear) (2)
Wherein, I is the image collected, Xnear, XfarIt is respectively that near and far cuts out plane；All object crawl examples pair are deleted, its
Middle image variance is less than 1e-3；When camera heights and table height match, all crawls of collected image are deleted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710141949.7A CN106886165A (en) | 2017-03-10 | 2017-03-10 | A kind of manipulator crawl and method of operating based on simulator |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710141949.7A CN106886165A (en) | 2017-03-10 | 2017-03-10 | A kind of manipulator crawl and method of operating based on simulator |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106886165A true CN106886165A (en) | 2017-06-23 |
Family
ID=59179603
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710141949.7A Pending CN106886165A (en) | 2017-03-10 | 2017-03-10 | A kind of manipulator crawl and method of operating based on simulator |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106886165A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109144036A (en) * | 2018-10-22 | 2019-01-04 | 江苏艾科半导体有限公司 | A kind of manipulator simulated testing system and test method based on fpga chip |
| CN110115849A (en) * | 2019-04-30 | 2019-08-13 | 厦门大学 | A kind of small-sized marionette robot control method, system, terminal device |
| WO2020016717A1 (en) * | 2018-07-19 | 2020-01-23 | International Business Machines Corporation | Perform peg-in-hole task with unknown tilt |
| CN111861305A (en) * | 2018-10-30 | 2020-10-30 | 牧今科技 | Robotic system with automated package registration mechanism and minimum feasible area detection |
| CN113043325A (en) * | 2019-12-27 | 2021-06-29 | 沈阳新松机器人自动化股份有限公司 | Method and device for detecting motion state of robot joint |
| US11780101B2 (en) | 2018-10-30 | 2023-10-10 | Mujin, Inc. | Automated package registration systems, devices, and methods |
-
2017
- 2017-03-10 CN CN201710141949.7A patent/CN106886165A/en active Pending
Non-Patent Citations (1)
| Title |
|---|
| MATTHEW VERES等: "An Integrated Simulator and Dataset that Combines Grasping and Vision for Deep Learning", 《ARXIV(HTTPS://ARXIV.ORG/ABS/1702.02103V1)》 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2588873B (en) * | 2018-07-19 | 2021-10-13 | Ibm | Perform peg-in-hole task with unknown tilt |
| CN112424703A (en) * | 2018-07-19 | 2021-02-26 | 国际商业机器公司 | Performing pin hole tasks with unknown inclinations |
| WO2020016717A1 (en) * | 2018-07-19 | 2020-01-23 | International Business Machines Corporation | Perform peg-in-hole task with unknown tilt |
| GB2588873A (en) * | 2018-07-19 | 2021-05-12 | Ibm | Perform peg-in-hole task with unknown tilt |
| US10953548B2 (en) | 2018-07-19 | 2021-03-23 | International Business Machines Corporation | Perform peg-in-hole task with unknown tilt |
| CN109144036A (en) * | 2018-10-22 | 2019-01-04 | 江苏艾科半导体有限公司 | A kind of manipulator simulated testing system and test method based on fpga chip |
| CN109144036B (en) * | 2018-10-22 | 2023-11-21 | 江苏艾科半导体有限公司 | Manipulator simulation test system and test method based on FPGA chip |
| US11288810B2 (en) | 2018-10-30 | 2022-03-29 | Mujin, Inc. | Robotic system with automated package registration mechanism and methods of operating the same |
| CN111861305A (en) * | 2018-10-30 | 2020-10-30 | 牧今科技 | Robotic system with automated package registration mechanism and minimum feasible area detection |
| US11501445B2 (en) | 2018-10-30 | 2022-11-15 | Mujin, Inc. | Robotic system with automated package scan and registration mechanism and methods of operating the same |
| US11636605B2 (en) | 2018-10-30 | 2023-04-25 | Mujin, Inc. | Robotic system with automated package registration mechanism and minimum viable region detection |
| US11780101B2 (en) | 2018-10-30 | 2023-10-10 | Mujin, Inc. | Automated package registration systems, devices, and methods |
| US11797926B2 (en) | 2018-10-30 | 2023-10-24 | Mujin, Inc. | Robotic system with automated object detection mechanism and methods of operating the same |
| US11961042B2 (en) | 2018-10-30 | 2024-04-16 | Mujin, Inc. | Robotic system with automated package registration mechanism and auto-detection pipeline |
| CN110115849A (en) * | 2019-04-30 | 2019-08-13 | 厦门大学 | A kind of small-sized marionette robot control method, system, terminal device |
| CN113043325B (en) * | 2019-12-27 | 2022-08-16 | 沈阳新松机器人自动化股份有限公司 | Method and device for detecting motion state of robot joint |
| CN113043325A (en) * | 2019-12-27 | 2021-06-29 | 沈阳新松机器人自动化股份有限公司 | Method and device for detecting motion state of robot joint |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106886165A (en) | A kind of manipulator crawl and method of operating based on simulator | |
| Newbury et al. | Deep learning approaches to grasp synthesis: A review | |
| Popović et al. | A strategy for grasping unknown objects based on co-planarity and colour information | |
| JP5743499B2 (en) | Image generating apparatus, image generating method, and program | |
| Kang et al. | Toward automatic robot instruction from perception-mapping human grasps to manipulator grasps | |
| Ekvall et al. | Learning and evaluation of the approach vector for automatic grasp generation and planning | |
| JP7022076B2 (en) | Image recognition processors and controllers for industrial equipment | |
| CN109015640B (en) | Grabbing method, grabbing system, computer device and readable storage medium | |
| JP5458885B2 (en) | Object detection method, object detection apparatus, and robot system | |
| CN110355754A (en) | Robot eye system, control method, equipment and storage medium | |
| Eizicovits et al. | Efficient sensory-grounded grasp pose quality mapping for gripper design and online grasp planning | |
| CN108712946A (en) | Cargo arrangement method, device, system and electronic equipment and readable storage medium storing program for executing | |
| Jiang et al. | Learning hardware agnostic grasps for a universal jamming gripper | |
| CN105196290B (en) | Real-time robot Grasp Planning | |
| WO2020190166A1 (en) | Method and system for grasping an object by means of a robotic device | |
| Moisio et al. | Model of tactile sensors using soft contacts and its application in robot grasping simulation | |
| CN109213202A (en) | Cargo arrangement method, device, equipment and storage medium based on optical servo | |
| Lepora et al. | Pose-based tactile servoing: Controlled soft touch using deep learning | |
| Rydén et al. | A method for constraint-based six degree-of-freedom haptic interaction with streaming point clouds | |
| Cao et al. | Fuzzy-depth objects grasping based on fsg algorithm and a soft robotic hand | |
| Song et al. | Learning optimal grasping posture of multi-fingered dexterous hands for unknown objects | |
| Caselli et al. | Haptic object recognition with a dextrous hand based on volumetric shape representations | |
| CN117103277A (en) | Mechanical arm sensing method based on multi-mode data fusion | |
| Marchionne et al. | GNC architecture solutions for robust operations of a free-floating space manipulator via image based visual servoing | |
| Kawasaki et al. | Virtual robot teaching for humanoid hand robot using muti-fingered haptic interface |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170623 |
|
| WD01 | Invention patent application deemed withdrawn after publication |