CN109108942A - The mechanical arm motion control method and system of the real-time teaching of view-based access control model and adaptive DMPS - Google Patents
The mechanical arm motion control method and system of the real-time teaching of view-based access control model and adaptive DMPS Download PDFInfo
- Publication number
- CN109108942A CN109108942A CN201811057825.1A CN201811057825A CN109108942A CN 109108942 A CN109108942 A CN 109108942A CN 201811057825 A CN201811057825 A CN 201811057825A CN 109108942 A CN109108942 A CN 109108942A
- Authority
- CN
- China
- Prior art keywords
- mechanical arm
- teaching
- motion
- dmps
- information
- 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.)
- Granted
Links
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
- B25J9/0081—Programme-controlled manipulators with master teach-in means
-
- 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
- B25J9/1605—Simulation of manipulator lay-out, design, modelling of manipulator
Abstract
The present invention relates to the mechanical arm motion control methods and system of a kind of real-time teaching of view-based access control model and adaptive DMPS.Teaching object is set first, then control teaching object carries out exemplary motion, depth map is obtained using Kinect and PnP algorithm is combined to carry out three-dimensional pose positioning and tracking to teaching object, it establishes space mapping system and teaching level appearance is mapped to mechanical arm tail end, it resolves each joint control information of mechanical arm according to inverse kinematics and sends in real time to indirectly control manipulator motion, final online record teaching motion information simultaneously carries out local linear optimization and study to it with adaptive DMPS algorithm.The present invention gets rid of the constraint with the dependence to sophisticated sensors of mechanical arm hardware configuration in traditional teaching mode, reduce the hardware cost and teaching difficulty of teaching, contactless characteristic also has broad applicability while enhancing the safety of teaching process, adaptive DMPS method proposed by the present invention makes whole system have good anti-interference.
Description
Technical field
The invention belongs to manipulator motion planning fields, and in particular to one kind is carried out based on QR code (one kind of two-dimensional bar code)
The online vision learning from instruction of mechanical arm motion control method and system.
Background technique
Motion planning is mainly for the mechanical arm module in robot with high-dimensional space, different from the road of plane
Diameter planning, is broadly divided into joint space trajectory planning and cartesian space trajectory planning two major classes, the former is main in tradition
With Spline Interpolation Method, the latter mainly uses the planing methods such as space line or space circular arc.Due to mechanical arm itself
Multiple degrees of freedom spatial character, so that not only planning calculates complexity greatly in these types of method practice, new also for each
Target will recalculate each posture of mechanical arm in motion process, especially when between mechanical arm and moving target point exist barrier
When hindering object, traditional method is more difficult to cook up the movement of mechanical arm, and it is complicated, intelligent low, suitable to there is calculating in the conventional way
The defects of answering property is poor.Open source motion planning library (the Open Motion Planning of Rice Univ USA Kavraki et al. exploitation
Library, OMPL) become current mainstream manipulator motion planning platform.It is suitable for the motion planning method of higher-dimension in OMPL
Mainly there is probability Lu Tufa (Probabilistic Road Map, PRM), quickly expand random tree method (Rapidly Random
Tree, RRT) and Artificial Potential Field Method etc., these algorithms use different sample mode regions of search, have planning speed fastly, generally
The advantages that rate is complete, but since the characteristic of algorithm stochastical sampling not only has so that the result that same task is planned every time is all different
When be unable to reach satisfied planning effect, can not also program results be made with anticipation.
In recent years, machine learning is quickly grown, and researchers are dedicated to developing the manipulator motion side of learning from instruction mode
Method allows the use of mechanical arm to become simple.How a great problem that on-line teaching is research and development, some research aircrafts are carried out to mechanical arm
Structure using customization can teaching type mechanical arm such as KUKA mechanical arm, such mechanical arm can tie up mechanical arm and user's arm
Surely make it while moving carry out on-line teaching, such methods not only need the hardware device of profession, and there is also one for teaching process
Fixed risk;Other research institutions are satisfied with user's motive position using a large amount of sensor and carry out teaching, this method
Not only the problems such as needing to rely on a large amount of and expensive sensors, but also being easy to appear interference influences learning effect, when reduction sensor number
It not can guarantee precision when amount, and have to artificially configure several auxiliary devices to increase precision.Thus how to implement one kind
Not only safety but also the extensive teaching method of applicability were the targets that industry is pursued always.
Summary of the invention
The present invention is for teaching difficulty in above-mentioned manipulator motion planning field learning from instruction mode and study anti-interference
The problems such as poor, proposes the mechanical arm motion control method and system of a kind of real-time teaching of view-based access control model and adaptive DMPS, the party
Method is greatly decreased number of sensors and guarantees precision, get rid of in traditional teaching mode the constraint of mechanical arm hardware configuration with to multiple
The dependence of miscellaneous sensor, reduces the hardware cost and teaching difficulty of teaching, and contactless characteristic enhances user's safety
Also there is broad applicability simultaneously, there is the problems such as interfering for teaching information, adaptive DMPS method proposed by the present invention makes
Whole system has good anti-interference.
In order to solve the above technical problems, the present invention adopts the following technical scheme:
A kind of mechanical arm motion control method of view-based access control model real-time teaching and adaptive DMPS, it is characterised in that pass through meter
Calculation machine vision carries out three-dimensional pose identification, positioning, tracking to the teaching object with setting QR code feature, and QR code center is two dimensional code
Teaching recognition site, two dimensional code teaching recognition site periphery are that white rectangle region connects to be formed with the small rectangle of the black of quadrangle
Rectangle frame;Space mapping system is established simultaneously by teaching object space information MAP to mechanical arm tail end, user passes through when teaching
Manipulate the indirect real-time control machinery arm movement of teaching object.
Further, the above method includes the following steps: to set teaching object first, carries out demonstration fortune by controlling teaching object
It is dynamic, obtain depth map using Kinect camera and teaching object identified in conjunction with PnP algorithm, three-dimensional pose positioning with
Track;It establishes space mapping system and teaching object space pose is mapped to mechanical arm tail end pose, according to inverse kinematics real-time resolving
It obtains each joint angle control information and sends and moved to the indirect real-time control machinery arm of mechanical arm, final online acquisition and recording teaching
Motion information simultaneously carries out local linear optimization and study to motion information with adaptive DMPS algorithm;The teaching object depth degree letter
Breath is mainly extracted from the depth map that Kinect depth camera obtains.
Further, online record teaching motion information and local linear optimization is carried out to it with adaptive DMPS algorithm
It is as follows with the process of study:
Teaching object space motion feature when online record training, first carries out Three Degree Of Freedom decomposition for the motion feature of record,
And learnt to obtain Optimal Nonlinear item warrant sequence with DMPS in each freedom degree, setting fresh target spatial information is simultaneously
It is extensive out to the motion feature of fresh target in each freedom degree;Extensive rear aimed at precision threshold value is set, extensive effect is learnt
Quality judges by the way that whether extensive result meets precision threshold, and the interference present in the teaching campaign causes to learn extensive result
When beyond precision threshold, it is excellent to will exceed the sample progress local least square method high order polynomial fitting progress that threshold value corresponds in freedom degree
Change, use DMPS study extensive again the sample motion feature after optimization, makes extensive result finally can be accurate extensive to new mesh
Punctuate;Mechanical arm tail end space fortune is fitted under same cannoncial system according to the extensive result of the Three Degree Of Freedom for finally meeting threshold value
Then dynamic feature controls manipulator motion by the motion information that inverse dynamics calculate each joint of mechanical arm.
In above-mentioned technical proposal, teaching motion information and mechanical arm command information at ROS real-time, interactive in following steps:
Step 1: building ROS environment, establish Kinect camera, mechanical arm information exchange node, correction Kinect camera shooting
Simultaneously image transmitting size and frequency is arranged in head parameter;The control module of mechanical arm is initialized, setting mechanical arm order receives and hair
Cloth frequency;
Step 2: setting with the object for setting QR code as teaching object, setting QR code is known using Kinect camera
, do not extract teaching object depth information in conjunction with recognition result and respective depth hum pattern, by setting QR code two-dimensional image information and
Depth information solves its spatial pose by PnP algorithm, while establishing three-dimensional pose coordinate system at teaching object QR code;For asking
4 known points of solution PnP algorithm are the central point of the peripheral four small rectangles of black of identification setting QR code;
Step 3: mechanical arm D-H Mo Xing is established, designs mechanical arm tail end relative to pedestal converting system according to positive kinematics,
Spatial pose then in conjunction with teaching object relative to Kinect establishes space mapping system;
Step 4: controlling teaching object before Kinect camera and carry out exemplary motion, teaching object in online record exemplary motion
Three-dimensional motion information;
Step 5: the movement of teaching object is mapped as by mechanical arm tail end movement according to the space mapping system established in step 3,
The motion information that each joint of mechanical arm is calculated by inverse dynamics sends joint motions to slave computer motion control card by node
Order driving mechanical arm real time kinematics, realize the real-time teaching of mechanical arm vision;
Step 6: the training motion information that step 4 records being decomposed into three one-dimensional movements on x, y, z Three Degree Of Freedom and is believed
Breath, single motion in one dimension information are the continuous time series of one group of displacement, velocity and acceleration
It indicates, takes step delta t by the motion information discretization, t ∈ { Δ t, 2 Δ t ..., n Δ t }, moves starting point x at this time0=xdemo
(0), terminal g=xdemo(0), run duration constant, τ=n Δ t, is utilized respectively DMPS algorithm to single freedom on three degree of freedom
Degree motion feature is learnt, and learning right repeated order column w is calculatedi;
Step 7: setting new moving target and Generalization accuracy threshold value, believed by the weight sequence and fresh target that learn to obtain
Extensive fresh target motion feature out is ceased, if the extensive result of certain single-degree-of-freedom is greater than precision threshold, to the instruction of corresponding freedom degree
Practice motion feature progress local least square method high order polynomial fitting to optimize, and the training after optimization is moved into return step 6
It relearns;
Step 8: obtained x, y, z three will be learnt in step 7 meet the fresh target motion in one dimension feature of required precision
It is fitted under same cannoncial system, obtains mechanical arm for the three-dimensional space motion information of fresh target, then use inverse fortune
Dynamic learn is decomposed, and the motion information in each joint of mechanical arm is resolved;
Step 9: the motion information in each joint of the resulting mechanical arm of step 8 being sent to slave computer motion control card, makes machine
Tool arm independently accurately moves to fresh target point according to motion information.
In above-mentioned technical proposal, in step 1, setting image size is 640 × 480, and it is 30 frame per second that image, which obtains frequency,
It is 30HZ that mechanical arm node orders, which are arranged, to issue and receive frequency.
In above-mentioned technical proposal, in step 2, teaching object location algorithm is PnP algorithm, and teaching object depth information is
QR code central depths are set in Kinect depth map;Positioning accuracy is 1mm.
In above-mentioned technical proposal, in step 3, tie point is Kinect and mechanical arm pedestal in installation space converting system,
I.e. teaching object is mechanical arm tail end relative to mechanical arm pedestal pose relative to Kinect pose.
In above-mentioned technical proposal, in step 7, it is 5% that extensive threshold accuracy, which is arranged, and training movement local optimum range is arranged
It is the 5%~95% of movement entirety, it is 8~15 ranks that polynomial order in least square higher order polynomial-fitting, which is arranged,.
In above-mentioned technical proposal, in step 9, it is 30HZ, joint of mechanical arm order letter that setting host computer order, which sends frequency,
Ceasing precision is 0.01 °.
A kind of system using the above method, characterized by comprising:
Surface has the teaching object that can be moved of setting QR code feature, and the teaching object is moved by demonstrator or autonomous fortune
It is dynamic;
Kinect camera, for identification QR code on teaching object;
Host computer communicates with Kinect camera and obtains teaching object motion information;
Mechanical arm equipped with end effector, mechanical arm and host computer are long-range or Near Field Communication is to follow teaching object to move
And it moves.
The present invention realizes the mechanical arm motion control method of a kind of view-based access control model real-time teaching and adaptive DMPS, i.e., logical
It crosses computer vision and there is the object of setting QR code to be identified setting, three-dimensional localization and tracking, pass through space conversion system
Control teaching object is set to carry out real-time control machinery arm end movement, and the motion feature of online acquisition teaching object is learnt.
Compared with the existing technology, the invention has the following advantages that
Teaching process proposed by the present invention does not need to reduce teaching using dedicated robotic arm apparatus and sophisticated sensors
Required hardware cost;
Teaching process proposed by the present invention is not constrained by mechanical arm hardware configuration, has wider applicability;
Using with the progress teaching of setting QR code object and teaching object and machine in the real-time teaching of vision proposed by the present invention
Tool arm greatly strengthens the safety of teaching process without directly contacting;
Adaptive DMPS proposed by the present invention, which solves in teaching campaign to exist to interfere, causes study Generalization accuracy is poor to ask
Topic, whole system have certain anti-interference ability.
The real-time teaching of entire vision proposed by the present invention and on-line study system make do not have robot field's relevant knowledge
Layman can simple and safely use mechanical arm, have ease for use and practicability;
Detailed description of the invention
Fig. 1 is the real-time teaching system structure chart of mechanical arm vision of the present invention.
Fig. 2 is the setting QR code embodiment for the three-dimensional localization for identification that the present invention designs.
Fig. 3 is the process of a kind of real-time teaching of view-based access control model of the present invention and the mechanical arm motion control method of adaptive DMPS
Figure.
Fig. 4 is the real-time teaching procedure chart of mechanical arm vision;Wherein (a)-(d) is followed successively by mechanical arm teaching object is followed to transport in real time
Dynamic different location figure.
Fig. 5 is teaching campaign Three Degree Of Freedom exploded orthogonal view;Wherein (a)-(c) is followed successively by the exploded view of tri- axis of xyz.
Fig. 6 is the extensive figure of fresh target three-degree-of-freedom motion feature;Wherein (a)-(c) is followed successively by the extensive figure of tri- axis of xyz.
Fig. 7 is comparison diagram before and after x freedom degree teaching motion optimization.
Fig. 8 is extensive comparative result figure before and after x freedom degree teaching motion optimization.
Specific embodiment
Technical solution in order to further illustrate the present invention, 1-8 is to a kind of view-based access control model of the invention with reference to the accompanying drawing
Real-time teaching and the mechanical arm motion control method and system of adaptive DMPS are described in detail.
It is as shown in Figure 1 the real-time teaching system structure chart of mechanical arm vision used by the method for the present invention.Wherein video camera 1
For reading the teaching object for having setting QR teaching code 2, the teaching object is static or motion state, and end effector is housed
Mechanical arm 3 and host computer is long-range or Near Field Communication, host computer obtain the representative of QR teaching code 2 by video camera 1 and show
Object depth degree and location information are taught, and then controls mechanical arm 3 and moves.
As shown in Fig. 2, the setting QR teaching code 2 on teaching object is rectangle two dimensional code, the rectangular color lumps 5 including rectangle quadrangle
And the two-dimension code area 4 between four rectangular color lumps 5.Wherein, rectangular color lumps 5 are that PNP algorithm positions position, two dimensional code
Region 4 is teaching recognition site.
The present invention is based on the mechanical arm motion control methods of vision real-time teaching and adaptive DMPS, as shown in figures 3-8, adopt
The real-time teaching object space three-dimensional localization of vision is carried out with Kinect and PnP algorithm to track, and establishes teaching object and mechanical arm actuator
Space reflection relationship enables user that can move by the indirect real-time control machinery arm of control teaching object, and can be by adaptively moving
State moves primitive algorithm (Dynamic Movement Primitives, hereinafter referred to as DMPS) learning training motion information, thus
Realize manipulator motion study with it is autonomous accurately extensive.
The present invention obtains depth map using Kinect and PnP algorithm is combined to carry out the teaching object with setting QR code feature
Three-dimensional pose locating and tracking, while establishing space mapping system for teaching object space information MAP to mechanical arm tail end, when teaching
User can be moved by the indirect real-time control machinery arm of manipulation teaching object, and Generalization accuracy threshold value is arranged by teaching motion information
It is put into and is learnt to obtain excellent non-linear convergence in adaptive DMPS as a result, in conjunction with the extensive accurate space out of mission bit stream
Motion information, last inverse kinematics resolve the motion information in each joint of mechanical arm, and motion information is sent to slave computer movement control
Fabrication can realize mechanical arm autokinetic movement.
It is set first with the object for setting QR code as teaching object, controls teaching object and carry out exemplary motion, pass through Kinect
Teaching object is identified with PnP (Perspective-n-Point, hereinafter referred to as PnP) algorithm, in conjunction with depth information to showing
Object is taught to carry out three-dimensional localization and tracking.It establishes space conversion system and teaching object space pose is mapped as mechanical arm tail end pose,
Go out the motion information real-time control machinery arm movement in each joint of mechanical arm according to inverse kinematics.Then when online record training
Its Three Degree Of Freedom is decomposed and is learnt to obtain with DMPS in each freedom degree best non-by teaching object space motion feature
The effect of linear term warrant sequence, study can be examined with to the extensive error of fresh target, and setting fresh target spatial information is simultaneously
It is extensive out to the motion feature of fresh target in each freedom degree, extensive rear aimed at precision threshold value is set, when due to teaching campaign
Present in interference cause to learn it is extensive after result beyond precision threshold when, will exceed threshold value and correspond to sample progress in freedom degree
The fitting of local least square method high order polynomial optimizes, and uses DMPS study extensive again the sample motion feature after optimization,
Make it finally can be accurate extensive to fresh target point.According to the extensive result of the Three Degree Of Freedom for finally meeting threshold value in same cannoncial system
Under be fitted to mechanical arm tail end spatial movement feature, then by inverse dynamics resolve each joint of mechanical arm motion information control machine
The movement of tool arm.
Above-mentioned technical proposal whole process flow chart is as shown in Fig. 1, and the specific implementation steps are as follows:
Step 1: building ROS environment, establish Kinect, mechanical arm information exchange node, correct Kinect camera parameter
And image transmitting size and frequency are set;The control module for initializing mechanical arm is arranged its order and receives and publication frequency.In detail
Steps are as follows:
Step 1-1: it is fixed that Kinect two dimension node of graph, Kinect depth node of graph, teaching object three-dimensional are established under ROS environment
Position node, space mapping system node, mechanical arm command node;
Step 1-2: it is 30 frame per second that setting Kinect image, which obtains frequency, and obtaining image size is 640 × 480, setting
The publication of mechanical arm node orders is 30HZ with frequency is received;
Step 1-3: setting slave computer motion control card serial port baud rate is 9600 (the mechanical arm selection forces that the present invention uses
Han Jiesheng Innovation Co., Ltd, model JS-R sixdegree-of-freedom simulation, motor are controlled using PWM), initialization mechanical arm respectively closes
Save posture;
Step 2: as shown in Fig. 2, the QR code of design setting of the present invention is shown as teaching object feature using Kinect acquisition
Object X-Y scheme and depth map information are taught, each frame image zooming-out of acquisition is identified to obtain four space of points poses, uses PnP
Algorithm positions its three-dimensional pose to teaching object feature QR code.Detailed step is as follows:
Step 2-1: obtaining teaching object X-Y scheme and depth map using Kinect, for acquisition each frame image recognition its
QR code is set on middle teaching object;
Step 2-2: four black rectangle central points of the peripheral rectangle of the QR code for taking identification to obtain are known point, and vision is arranged
World coordinate system is Kinect coordinate system in system, in X-Y scheme location of pixels be (u, v) vertex it is corresponding its in vision
Three-dimensional coordinate (X under systemc, Yc, Zc) are as follows:
Here present invention calibration Kinect internal reference is f=(fx, fy, u0, v0)=(525,525,319.5,239.5), by
Kinect depth map obtains each vertex depth Zc;
Step 2-3: by four known point (Xwi, Ywi, Zwi), i=1,2,3,4 find out teaching object rotates in vision system
Matrix RvisionWith translation vector TvisionTo get teaching level appearance is arrived, specific solution is as follows:
Here TzFor teaching object mean depth, middle setting setting QR code is symmetrical rectangular according to the present invention, takes TzFor in rectangle
The heart, that is, QR code central depths;
Step 3: mechanical arm D-H Mo Xing is established, designs mechanical arm tail end relative to pedestal converting system according to positive kinematics,
Spatial pose then in conjunction with teaching object in vision system establishes space mapping system.The present invention is arranged the world in vision system and sits
Mark system is Kinect coordinate system, and it is mechanical arm base coordinate system that world coordinate system in mechanical arm system, which is arranged, by preferred,
Mechanical arm tail end posture under installation space mapped system are as follows:
Step 4: controlling teaching object before Kinect and carry out spatial movement, teaching object motion bit is positioned according to step 2 in real time
It sets and posture, while teaching object three-dimensional motion information in online record training movement;
Step 5: as shown in Fig. 3, the movement of teaching object being mapped as by machinery according to the space mapping system established in step 3
Arm end movement is resolved the motion information in each joint of mechanical arm by inverse dynamics, as shown in Fig. 4, by node to slave computer
Motion control card sends joint motions order driving mechanical arm and follows teaching object real time kinematics, realizes that mechanical arm vision is shown in real time
Religion;(a) of Fig. 4-(d) is successively
Step 6: the training motion information that step 4 records is decomposed into three one-dimensional movements on x, y, z Three Degree Of Freedom
Information, single motion in one dimension information are the continuous sequence of one group of displacement, velocity and acceleration
It indicates, takes step delta t that will move discretization, at this time t ∈ { Δ t, 2 Δ t ..., n Δ t }, move starting point x0=xdemo(0), terminal g
=xdemo(0), run duration constant, τ is utilized respectively DMPS algorithm to single dof mobility feature in three freedom
It practises, calculates learning right repeated order column wi, detailed step is as follows:
Step 6-1: as shown in Fig. 5, the teaching object for the vision servant control that step 4 is recorded is moved in x, y, z three
Orthogonal Decomposition in a freedom degree obtains three one-dimensional motion informations, and it is discrete that the present invention takes step delta t=0.05 to move one-dimensional
Change, timeconstantτ=20
Step 6-2: it is utilized respectively DMPS algorithm in three freedom, the single dimension motion feature that step 6-1 is obtained is carried out
Study, by the nonlinear function f of learning objectivetarget(s) the nonlinear function f of training movement is approacheddemo(s), pass through minimum
Two multiply error of fitting Ji, learn to obtain convergent weight sequence wi, the weight number that weight sequence is arranged in the present invention is 10;
Step 7: setting new moving target and Generalization accuracy threshold value, pass through the weight sequence and fresh target information for learning to obtain
Extensive fresh target motion feature out, if the extensive result of certain single-degree-of-freedom is greater than precision threshold, the training to corresponding freedom degree
Motion feature carries out the fitting removal disturbance of local least square method high order polynomial, and will train movement 6 weight of return step after optimization
New study.Detailed step is as follows:
Step 7-1: setting new moving target and Generalization accuracy threshold value, and fresh target is arranged compared to original trained mesh in the present invention
Mark Three Degree Of Freedom position is different, and Three Degree Of Freedom Generalization accuracy threshold value is all 5%, specifically:
(xnew, ynew, znew)=(xtraining+ 0.1m, ytraining+ 0.1m, ztraining+0.1m)
Step 7-2: as shown in Fig. 5, according to the step 6-2 weight sequence learnt and step 7-1 fresh target information
The motion feature of fresh target, then finds out the extensive result precision of each freedom degree on extensive Three Degree Of Freedom out, and precision is arranged in the present invention
It solves as follows:
Generalization accuracy=| (the extensive extensive target of result -)/(fresh target-training objective) |
I.e. Generalization accuracy is that extensive result absolute error takes absolutely divided by the value of coordinates of targets variable quantity in corresponding freedom degree
Value, x, y, z Three Degree Of Freedom Generalization accuracy of the present invention are each are as follows: 10.87%, 5.64%, 7.66%;
Step 7-3: exceeding precision threshold for Generalization accuracy obtained by step 7-2 and the comparison of the Generalization accuracy 5% of setting, with
For x freedom degree, as shown in Fig. 7, to by teaching campaign continuous sequence in x freedom degree
5%~95% whole part carries out the fitting optimization of least square high order polynomial, the x freedom degree motion information sequence after being optimized
Column, the present invention are arranged movement in x freedom and carry out 8 rank multinomial optimizations;
Step 7-4: motion information in the x freedom degree after optimizing in step 7-3 is re-started study by return step 6, with
It is extensive that step 7 is carried out afterwards, and extensive result is as shown in Fig. 8 after present invention optimization, and extensive result precision is after the optimization of x freedom degree
3%, it is less than extensive threshold value 5%, meets the requirements;
Step 7-5: with x freedom degree Optimization Steps, y, z freedom degree for being unsatisfactory for required precision are moved optimize until
All meet Generalization accuracy;
Step 8: by three fresh target motion in one dimension features of resulting x, y, z extensive in step 7 under same cannoncial system
It is fitted, obtains fresh target three-dimensional space motion information, then decomposed with inverse kinematics, obtain each pass of mechanical arm
The motion information of section;
Step 9: the motion information in each joint of the resulting mechanical arm of step 8 being sent to slave computer motion control card, is made
Mechanical arm independently accurately moves to fresh target point according to motion information.
In above-mentioned technical proposal, the teaching object is characterized by setting QR (Quick Response, hereinafter referred to as QR)
The object of code is identified as teaching object to identify that it sets QR code;
In above-mentioned technical proposal, the teaching object depth information is mainly in the depth map that Kinect depth camera obtains
It extracts;
In above-mentioned technical proposal, the teaching object motion information and mechanical arm command information real-time, interactive at ROS.
In above-mentioned technical proposal, in step 1, setting image size is 640 × 480, and it is 30 frame per second that image, which obtains frequency,
It is 30HZ that mechanical arm node orders, which are arranged, to issue and receive frequency;
In above-mentioned technical proposal, in step 2, teaching object feature are as follows: teaching object has setting QR code, outer centered on QR code
Enclose by quadrangle be the small rectangle of black white area surround, whole image having a size of 8cm × 8cm, center QR code having a size of 5cm ×
5cm rectangle, the small rectangular dimension of black are 1.5cm × 1.5cm, and teaching object location algorithm is PnP (Perspective-n-Point)
Algorithm, positioning accuracy 1mm, teaching object depth information are QR code central depths in Kinect depth map;
In above-mentioned technical proposal, in step 2, teaching object three-dimensional pose vision positioning is specific as follows:
Teaching level appearance has spin matrix R and translation vector T:
Teaching object is in internal reference f=(fx, fy, u0, v0) perspective projection transformation under Kinect are as follows:
Teaching object location of pixels is two dimensional image coordinate system coordinate corresponding to (u, v) in X-Y scheme are as follows:
For 4 spatial point (X known under world coordinate systemwi, Ywi, zwi), i=1,2,3,4 have:
Here TzFor teaching object mean depth, above 8 independent equations solve R1, R2, T is that orthogonal matrix solution goes out according to R
R3, so far solve teaching object whole pose.It is Kinect coordinate system that world coordinate system in vision system, which is arranged, in the present invention, and setting is reflected
World coordinate system is mechanical arm base coordinate system in the mechanical arm system penetrated.4 for solving PnP algorithm are set in the present invention
Known point is that identification setting QR code obtains black four central points of small rectangle.
In above-mentioned technical proposal, in step 3, setting converting system interior joint is Kinect and mechanical arm pedestal, i.e. teaching
Object is mechanical arm tail end relative to mechanical arm pedestal pose relative to Kinect pose;
In above-mentioned technical proposal, in step 6, learning process is specific as follows:
The core of dynamic motion basic-element theory is to describe to move by a series of nonlinear differential equations with target point.
For its expression of single motion in one dimension are as follows:
ψi(s)=exp (- hi(s-ci)2)
τ is time contraction-expansion factor, x in formula0It is system Origin And Destination state respectively with g, x and v are system current state
With speed, K, D are system constants, and f is the linear weighted function and ψ of radial basis functioniIt (s) is Radial basis kernel function, hiWith ciFor base
Kernel function bandwidth and mean value, N are the number of Radial basis kernel function, and w is weighted value of the Radial basis kernel function in linear weighted function, s
It is the function about time t, specific dynamic characteristic is defined by following equation:
This second-order system is cannoncial system, and α is predetermined constant (a > 0) here, and original state s (0)=1, s is by initial
State trend is in 0 process
It puts samples into and f can be obtained in DMPSdemo(s):
The process of motor learning, i.e., by nonlinear function fdemo(s) nonlinear function of authentic specimen model is approached
ftarget(s), fdemo(s) s is determined by cannoncial system in, restrains wiValue complete ftarget(s)→fdemo(s) it approaches.With minimum
Two multiply learning method error of fitting J:
J=∑S(fdemo(s)-ftarget(s))2
In J=JminWhen estimate weighted value wiAs system best weights weight values, estimated form:
F=Tw
W=[w1 …wN]T
When giving new moving target point gnew, according to best weights weight values w hereiNew motion feature can reversely be fitted
Discrete seriesIt is special that this discrete series can fit the movement of the fresh target with original sample eigen
Sign completes the extensive of movement.
In above-mentioned technical proposal, in step 7, it is 5% that extensive threshold accuracy, which is arranged, and training movement local optimum range is arranged
It is the 5%~95% of movement entirety, it is 8~15 ranks that polynomial order in least square higher order polynomial-fitting, which is arranged,;
In above-mentioned technical proposal, in step 9, it is 30HZ, joint of mechanical arm order letter that setting host computer order, which sends frequency,
Ceasing precision is 0.01 °;
To sum up, the manipulator motion learning method of a kind of real-time teaching of view-based access control model and adaptive DMPS proposed by the present invention
The teaching of mechanical arm real-time vision and accurate extensive function are realized, the real-time tutorial function of the vision that the present invention realizes has extensive
Applicability, the real-time teaching characteristic of vision not only reduce hardware cost needed for teaching, also greatly strengthen the peace of teaching process
Quan Xing, while this invention also solves final extensive result low precision is led to by interference in teaching campaign, it can accurately advise
Manipulator motion is drawn to target point, enhances the anti-interference of whole system.
Claims (10)
1. a kind of mechanical arm motion control method of view-based access control model real-time teaching and adaptive DMPS, it is characterised in that pass through calculating
Machine vision carries out three-dimensional pose identification, positioning, tracking to the teaching object with setting QR code feature, and QR code center shows for two dimensional code
Recognition site is taught, two dimensional code teaching recognition site periphery connects the square to be formed with the small rectangle of the black of quadrangle for white rectangle region
Shape frame;Space mapping system is established simultaneously by teaching object space information MAP to mechanical arm tail end, user passes through behaviour when teaching
Control the indirect real-time control machinery arm movement of teaching object.
2. the mechanical arm motion control method of view-based access control model real-time teaching and adaptive DMPS according to claim 1,
It is characterized in that the above method includes the following steps: to set teaching object first, exemplary motion is carried out by control teaching object, is used
Kinect camera obtains depth map and is identified in conjunction with PnP algorithm to teaching object, three-dimensional pose positions and tracking;It establishes empty
Between mapped system teaching object space pose is mapped to mechanical arm tail end pose, each joint is obtained according to inverse kinematics real-time resolving
Angle controls information and sends and moves to the indirect real-time control machinery arm of mechanical arm, and final online acquisition and recording teaching motion information is simultaneously
Local linear optimization and study are carried out to motion information with adaptive DMPS algorithm;The teaching object depth information mainly from
It is extracted in the depth map that Kinect depth camera obtains.
3. the mechanical arm motion control method of view-based access control model real-time teaching and adaptive DMPS according to claim 1,
It is characterized in that online record teaching motion information and carries out local linear optimization and study to it with adaptive DMPS algorithm
Process is as follows:
The motion feature of record is first carried out Three Degree Of Freedom decomposition by teaching object space motion feature when online record training, and
Learnt to obtain Optimal Nonlinear item warrant sequence with DMPS in each freedom degree, sets fresh target spatial information and every
It is extensive out to the motion feature of fresh target in a freedom degree;Extensive rear aimed at precision threshold value is set, the quality of extensive effect is learnt
Judge by the way that whether extensive result meets precision threshold, the interference present in the teaching campaign causes to learn extensive result to exceed
When precision threshold, it will exceed the sample progress local least square method high order polynomial fitting that threshold value corresponds in freedom degree and optimize,
Again it uses DMPS study extensive sample motion feature after optimization, makes extensive result finally can be accurate extensive to fresh target
Point;Mechanical arm tail end spatial movement is fitted under same cannoncial system according to the extensive result of the Three Degree Of Freedom for finally meeting threshold value
Then feature controls manipulator motion by the motion information that inverse dynamics calculate each joint of mechanical arm.
4. the mechanical arm motion control method of view-based access control model real-time teaching and adaptive DMPS according to claim 1,
Be characterized in that teaching motion information and mechanical arm command information at ROS real-time, interactive in following steps:
Step 1: building ROS environment, establish Kinect camera, mechanical arm information exchange node, correction Kinect camera ginseng
It counts and image transmitting size and frequency is set;The control module of mechanical arm is initialized, setting mechanical arm order receives and publication frequency
Rate;
Step 2: it sets with the object for setting QR code as teaching object, setting QR code is identified using Kinect camera,
Teaching object depth information is extracted in conjunction with recognition result and respective depth hum pattern, passes through setting QR code two-dimensional image information and depth
Information solves its spatial pose by PnP algorithm, while establishing three-dimensional pose coordinate system at teaching object QR code;For solving
4 known points of PnP algorithm are the central point of the peripheral four small rectangles of black of identification setting QR code;
Step 3: establishing mechanical arm D-H Mo Xing, design mechanical arm tail end relative to pedestal converting system, then according to positive kinematics
Spatial pose in conjunction with teaching object relative to Kinect establishes space mapping system;
Step 4: controlling teaching object before Kinect camera and carry out exemplary motion, teaching object is three-dimensional in online record exemplary motion
Motion information;
Step 5: the movement of teaching object being mapped as by mechanical arm tail end movement according to the space mapping system established in step 3, by inverse
Dynamics calculates the motion information in each joint of mechanical arm, sends joint motions order to slave computer motion control card by node
Mechanical arm real time kinematics are driven, realize the real-time teaching of mechanical arm vision;
Step 6: the training motion information that step 4 records is decomposed into three one-dimensional motion informations on x, y, z Three Degree Of Freedom,
Single motion in one dimension information is the continuous time series of one group of displacement, velocity and acceleration
It indicates, takes step delta t by the motion information discretization, t ∈ { Δ t, 2 Δ t ..., n Δ t }, moves starting point x at this time0=xdemo
(0), terminal g=xdemo(0), run duration constant, τ=n Δ t, is utilized respectively DMPS algorithm to single freedom on three degree of freedom
Degree motion feature is learnt, and learning right repeated order column w is calculatedi;
Step 7: new moving target and Generalization accuracy threshold value are set, it is general by the weight sequence and fresh target information that learn to obtain
Fresh target motion feature is dissolved, if the extensive result of certain single-degree-of-freedom is greater than precision threshold, the training of corresponding freedom degree is transported
Dynamic feature carries out the fitting of local least square method high order polynomial and optimizes, and again by the training movement return step 6 after optimization
Study;
Step 8: obtained x, y, z three will be learnt in step 7 and meet the fresh target motion in one dimension feature of required precision same
It is fitted under one cannoncial system, obtains mechanical arm for the three-dimensional space motion information of fresh target, then use inverse kinematics
It is decomposed, resolves the motion information in each joint of mechanical arm;
Step 9: the motion information in each joint of the resulting mechanical arm of step 8 being sent to slave computer motion control card, makes mechanical arm
Fresh target point is independently accurately moved to according to motion information.
5. the mechanical arm motion control method of view-based access control model real-time teaching and adaptive DMPS according to claim 1,
It is characterized in that, in step 1, setting image size is 640 × 480, and it is 30 frame per second that image, which obtains frequency, and mechanical arm node is arranged
Order publication is 30HZ with frequency is received.
6. the mechanical arm motion control method of view-based access control model real-time teaching and adaptive DMPS according to claim 1,
It is characterized in that, in step 2, teaching object location algorithm is PnP algorithm, and teaching object depth information is to set in Kinect depth map
QR code central depths;Positioning accuracy is 1mm.
7. the mechanical arm motion control method of view-based access control model real-time teaching and adaptive DMPS according to claim 1,
Be characterized in that, in step 3, tie point is Kinect and mechanical arm pedestal in installation space converting system, i.e., teaching object relative to
Kinect pose is mechanical arm tail end relative to mechanical arm pedestal pose.
8. the mechanical arm motion control method of view-based access control model real-time teaching and adaptive DMPS according to claim 1,
It is characterized in that, in step 7, it is 5% that extensive threshold accuracy, which is arranged, and it is that movement is whole that training movement local optimum range, which is arranged,
5%~95%, it is 8~15 ranks that polynomial order in least square higher order polynomial-fitting, which is arranged,.
9. the mechanical arm motion control method of view-based access control model real-time teaching and adaptive DMPS according to claim 1,
It is characterized in that, in step 9, it is 30HZ that setting host computer order, which sends frequency, and joint of mechanical arm command information precision is 0.01 °.
10. a kind of system using one of the claims 1-9 the method, characterized by comprising:
Surface has the teaching object that can be moved of setting QR code feature, and the teaching object is moved by demonstrator or autokinetic movement;
Kinect camera, for identification QR code on teaching object;
Host computer communicates with Kinect camera and obtains teaching object motion information;
Mechanical arm equipped with end effector, mechanical arm and host computer are long-range or Near Field Communication is transported with following teaching object to move
It is dynamic.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811057825.1A CN109108942B (en) | 2018-09-11 | 2018-09-11 | Mechanical arm motion control method and system based on visual real-time teaching and adaptive DMPS |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811057825.1A CN109108942B (en) | 2018-09-11 | 2018-09-11 | Mechanical arm motion control method and system based on visual real-time teaching and adaptive DMPS |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109108942A true CN109108942A (en) | 2019-01-01 |
CN109108942B CN109108942B (en) | 2021-03-02 |
Family
ID=64859196
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811057825.1A Active CN109108942B (en) | 2018-09-11 | 2018-09-11 | Mechanical arm motion control method and system based on visual real-time teaching and adaptive DMPS |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109108942B (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110026987A (en) * | 2019-05-28 | 2019-07-19 | 广东工业大学 | Generation method, device, equipment and the storage medium of a kind of mechanical arm crawl track |
CN110116116A (en) * | 2019-05-14 | 2019-08-13 | 中国地质大学(武汉) | Robotic laser cleaning path planning system based on computer vision and method |
CN110298854A (en) * | 2019-05-17 | 2019-10-01 | 同济大学 | The snakelike arm co-located method of flight based on online adaptive and monocular vision |
CN110471281A (en) * | 2019-07-30 | 2019-11-19 | 南京航空航天大学 | A kind of the Varied scope fuzzy control system and control method of Trajectory Tracking Control |
CN110481029A (en) * | 2019-09-05 | 2019-11-22 | 南京信息职业技术学院 | A kind of the 3D printing warpage preventing temperature compensation system and compensation method of position follower |
CN111002289A (en) * | 2019-11-25 | 2020-04-14 | 华中科技大学 | Robot online teaching method and device, terminal device and storage medium |
CN111216124A (en) * | 2019-12-02 | 2020-06-02 | 广东技术师范大学 | Robot vision guiding method and device based on integration of global vision and local vision |
CN111823215A (en) * | 2020-06-08 | 2020-10-27 | 深圳市越疆科技有限公司 | Synchronous control method and device for industrial robot |
CN111890353A (en) * | 2020-06-24 | 2020-11-06 | 深圳市越疆科技有限公司 | Robot teaching track reproduction method and device and computer readable storage medium |
CN112008692A (en) * | 2019-05-31 | 2020-12-01 | 精工爱普生株式会社 | Teaching method |
CN112207835A (en) * | 2020-09-18 | 2021-01-12 | 浙江大学 | Method for realizing double-arm cooperative work task based on teaching learning |
CN112476489A (en) * | 2020-11-13 | 2021-03-12 | 哈尔滨工业大学(深圳) | Flexible mechanical arm synchronous measurement method and system based on natural characteristics |
CN112530267A (en) * | 2020-12-17 | 2021-03-19 | 河北工业大学 | Intelligent mechanical arm teaching method based on computer vision and application |
CN112975975A (en) * | 2021-03-02 | 2021-06-18 | 路邦康建有限公司 | Robot control interface correction method and hospital clinical auxiliary robot thereof |
CN113977580A (en) * | 2021-10-29 | 2022-01-28 | 浙江工业大学 | Mechanical arm simulation learning method based on dynamic motion primitives and adaptive control |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030180697A1 (en) * | 2002-03-22 | 2003-09-25 | Kim Kyung Hwan | Multi-degree of freedom telerobotic system for micro assembly |
JP2010179403A (en) * | 2009-02-05 | 2010-08-19 | Denso Wave Inc | Robot simulation image display system |
EP2366502A1 (en) * | 2010-02-26 | 2011-09-21 | Honda Research Institute Europe GmbH | Robot with hand-object movement correlations for online temporal segmentation of movement tasks |
CN106142092A (en) * | 2016-07-26 | 2016-11-23 | 张扬 | A kind of method robot being carried out teaching based on stereovision technique |
CN106444738A (en) * | 2016-05-24 | 2017-02-22 | 武汉科技大学 | Mobile robot path planning method based on dynamic motion primitive learning model |
CN107160364A (en) * | 2017-06-07 | 2017-09-15 | 华南理工大学 | A kind of industrial robot teaching system and method based on machine vision |
-
2018
- 2018-09-11 CN CN201811057825.1A patent/CN109108942B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030180697A1 (en) * | 2002-03-22 | 2003-09-25 | Kim Kyung Hwan | Multi-degree of freedom telerobotic system for micro assembly |
JP2010179403A (en) * | 2009-02-05 | 2010-08-19 | Denso Wave Inc | Robot simulation image display system |
EP2366502A1 (en) * | 2010-02-26 | 2011-09-21 | Honda Research Institute Europe GmbH | Robot with hand-object movement correlations for online temporal segmentation of movement tasks |
CN106444738A (en) * | 2016-05-24 | 2017-02-22 | 武汉科技大学 | Mobile robot path planning method based on dynamic motion primitive learning model |
CN106142092A (en) * | 2016-07-26 | 2016-11-23 | 张扬 | A kind of method robot being carried out teaching based on stereovision technique |
CN107160364A (en) * | 2017-06-07 | 2017-09-15 | 华南理工大学 | A kind of industrial robot teaching system and method based on machine vision |
Non-Patent Citations (1)
Title |
---|
UDE ET AL.: "Task-Specific Generalization of Discrete and Periodic Dynamic Movement Primitives", 《IEEE TRANSACTIONS ON ROBOTICS》 * |
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110116116A (en) * | 2019-05-14 | 2019-08-13 | 中国地质大学(武汉) | Robotic laser cleaning path planning system based on computer vision and method |
CN110298854A (en) * | 2019-05-17 | 2019-10-01 | 同济大学 | The snakelike arm co-located method of flight based on online adaptive and monocular vision |
CN110298854B (en) * | 2019-05-17 | 2021-05-11 | 同济大学 | Flight snake-shaped arm cooperative positioning method based on online self-adaption and monocular vision |
CN110026987A (en) * | 2019-05-28 | 2019-07-19 | 广东工业大学 | Generation method, device, equipment and the storage medium of a kind of mechanical arm crawl track |
CN110026987B (en) * | 2019-05-28 | 2022-04-19 | 广东工业大学 | Method, device and equipment for generating grabbing track of mechanical arm and storage medium |
CN112008692A (en) * | 2019-05-31 | 2020-12-01 | 精工爱普生株式会社 | Teaching method |
CN110471281A (en) * | 2019-07-30 | 2019-11-19 | 南京航空航天大学 | A kind of the Varied scope fuzzy control system and control method of Trajectory Tracking Control |
CN110471281B (en) * | 2019-07-30 | 2021-09-24 | 南京航空航天大学 | Variable-discourse-domain fuzzy control system and control method for trajectory tracking control |
CN110481029A (en) * | 2019-09-05 | 2019-11-22 | 南京信息职业技术学院 | A kind of the 3D printing warpage preventing temperature compensation system and compensation method of position follower |
CN110481029B (en) * | 2019-09-05 | 2021-08-20 | 南京信息职业技术学院 | Position-follow-up 3D printing warping-prevention temperature compensation system and compensation method |
CN111002289A (en) * | 2019-11-25 | 2020-04-14 | 华中科技大学 | Robot online teaching method and device, terminal device and storage medium |
CN111002289B (en) * | 2019-11-25 | 2021-08-17 | 华中科技大学 | Robot online teaching method and device, terminal device and storage medium |
CN111216124B (en) * | 2019-12-02 | 2020-11-06 | 广东技术师范大学 | Robot vision guiding method and device based on integration of global vision and local vision |
CN111216124A (en) * | 2019-12-02 | 2020-06-02 | 广东技术师范大学 | Robot vision guiding method and device based on integration of global vision and local vision |
CN111823215A (en) * | 2020-06-08 | 2020-10-27 | 深圳市越疆科技有限公司 | Synchronous control method and device for industrial robot |
CN111890353A (en) * | 2020-06-24 | 2020-11-06 | 深圳市越疆科技有限公司 | Robot teaching track reproduction method and device and computer readable storage medium |
CN112207835A (en) * | 2020-09-18 | 2021-01-12 | 浙江大学 | Method for realizing double-arm cooperative work task based on teaching learning |
CN112476489A (en) * | 2020-11-13 | 2021-03-12 | 哈尔滨工业大学(深圳) | Flexible mechanical arm synchronous measurement method and system based on natural characteristics |
CN112476489B (en) * | 2020-11-13 | 2021-10-22 | 哈尔滨工业大学(深圳) | Flexible mechanical arm synchronous measurement method and system based on natural characteristics |
CN112530267A (en) * | 2020-12-17 | 2021-03-19 | 河北工业大学 | Intelligent mechanical arm teaching method based on computer vision and application |
CN112975975A (en) * | 2021-03-02 | 2021-06-18 | 路邦康建有限公司 | Robot control interface correction method and hospital clinical auxiliary robot thereof |
CN113977580A (en) * | 2021-10-29 | 2022-01-28 | 浙江工业大学 | Mechanical arm simulation learning method based on dynamic motion primitives and adaptive control |
Also Published As
Publication number | Publication date |
---|---|
CN109108942B (en) | 2021-03-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109108942A (en) | The mechanical arm motion control method and system of the real-time teaching of view-based access control model and adaptive DMPS | |
CN111203878B (en) | Robot sequence task learning method based on visual simulation | |
Billard et al. | Discriminative and adaptive imitation in uni-manual and bi-manual tasks | |
Billard et al. | Discovering optimal imitation strategies | |
Silvério et al. | Learning task priorities from demonstrations | |
Muhlig et al. | Task-level imitation learning using variance-based movement optimization | |
US11694432B2 (en) | System and method for augmenting a visual output from a robotic device | |
CN111872934B (en) | Mechanical arm control method and system based on hidden semi-Markov model | |
Delhaisse et al. | Transfer learning of shared latent spaces between robots with similar kinematic structure | |
Hueser et al. | Learning of demonstrated grasping skills by stereoscopic tracking of human head configuration | |
CN104850120A (en) | Wheel type mobile robot navigation method based on IHDR self-learning frame | |
CN115990891B (en) | Robot reinforcement learning assembly method based on visual teaching and virtual-actual migration | |
Zorina et al. | Learning to manipulate tools by aligning simulation to video demonstration | |
Vidaković et al. | Learning from demonstration based on a classification of task parameters and trajectory optimization | |
Liu et al. | Learning articulated constraints from a one-shot demonstration for robot manipulation planning | |
Mavsar et al. | Simulation-aided handover prediction from video using recurrent image-to-motion networks | |
Schwab | Robot Deep Reinforcement Learning: Tensor State-Action Spaces and Auxiliary Task Learning with Multiple State Representations. | |
Zhu | Robot Learning Assembly Tasks from Human Demonstrations | |
Xu et al. | Vision-Based Intelligent Perceiving and Planning System of a 7-DoF Collaborative Robot | |
Hüser et al. | Visual programming by demonstration of grasping skills in the context of a mobile service robot using 1D-topology based self-organizing-maps | |
Al-Shanoon | Developing a mobile manipulation system to handle unknown and unstructured objects | |
Zhang et al. | Learning Descriptor of Constrained Task from Demonstration | |
Qi et al. | Revolutionizing Packaging: A Robotic Bagging Pipeline with Constraint-aware Structure-of-Interest Planning | |
Tekden | Data Efficient Representation Learning for Grasping and Manipulation | |
Omrcen et al. | Sensorimotor processes for learning object representations |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |