CN108196453A - A kind of manipulator motion planning Swarm Intelligent Computation method - Google Patents
A kind of manipulator motion planning Swarm Intelligent Computation method Download PDFInfo
- Publication number
- CN108196453A CN108196453A CN201810068243.7A CN201810068243A CN108196453A CN 108196453 A CN108196453 A CN 108196453A CN 201810068243 A CN201810068243 A CN 201810068243A CN 108196453 A CN108196453 A CN 108196453A
- Authority
- CN
- China
- Prior art keywords
- mechanical arm
- crawl
- path
- value
- model
- 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
-
- 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
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of manipulator motion planning Swarm Intelligent Computation methods, include the following steps：Step 1：The three-dimensional system of coordinate of working space where building mechanical arm；Step 2：Build the object target identification model to be captured based on fuzzy neural network；Step 3：Build the mechanical arm crawl optimal path model based on extreme learning machine；Step 4：Object target image identification object classification to be captured is acquired in real time；Step 5；Object center of gravity is determined based on object classification, acquires crawl terminal point coordinate, obtains mechanical arm crawl optimal path, driving mechanical arm crawl object；This method is using the ideal trajectory that one collisionless of mechanical arm crawl target object is calculated in intelligent algorithm, kinetic characteristics meet margin requirement, path length and run duration are shorter, substantially increase production efficiency, personnel cost has been saved simultaneously, income is brought to factory.
Description
Technical field
The invention belongs to mechanical arms to control calculating field, more particularly to a kind of manipulator motion planning Swarm Intelligent Computation side
Method.
Background technology
As the arriving of " made in China 2025 ", intelligent robot such as manipulator, AGV, unmanned plane etc. will be in manufacturing environments
Under undertake more and more work.Existing machinery arm can be by teaching machine, PC control, and teaching machine control is led on teaching machine
It crosses control stick and button to control each joint of mechanical arm, then completes program storage after set operating mode；PC control
It is to be programmed in host computer, set Operation mode cycle can be completed.Above two Mechanical arm control method all can be only done fixation
The cycle of operating mode can not cope with the variation of operating mode, for existing machinery arm operation by the monitoring device operating robotic arm visual field by
Limit, operator easily generate during entire control and are difficult to the biology born fatigue, cause error that safety accident occurs, be
This, the automatic identification and crawl articles path design of mechanical arm are to solve a quantum jump of mechanical arm control problem.
Invention content
The present invention proposes a kind of manipulator motion planning Swarm Intelligent Computation method, and it is external that this method passes through mechanical arm
Camera establishes cartesian coordinate system and obtains target object image and identify target object, is obtained and captured by machine learning model
Target object optimal motion path, and barrier is judged, generation avoidance path.
A kind of manipulator motion planning Swarm Intelligent Computation method method, includes the following steps：
Step 1：The three-dimensional system of coordinate of working space where building mechanical arm；
Utilize a left side for the binocular ZED cameras of the entire working space of image acquisition region covering mechanical arm crawl target object
Right camera line midpoint as origin, using the right camera centers of binocular ZED to the left camera lines of centres of binocular ZED as y-axis
Forward direction establishes the 3 D stereo coordinate system of mechanical arm working space according to the right-hand rule；
Step 2：Build the object target identification model to be captured based on fuzzy neural network；
Using the intermediate pixel accumulated value of every width object target image to be captured and corresponding object classification as input
And output data, fuzzy neural network is trained, obtains object target the to be captured identification mould based on fuzzy neural network
Type；
The acquisition process of the intermediate pixel accumulated value of the object target image to be captured is as follows：
All kinds of object target images to be captured are acquired using the binocular ZED for being used to build 3 D stereo coordinate system in step 1
Image under various poses；It treats crawl object target image and carries out denoising, evolution, gray proces and edge inspection successively
It surveys；The size for extracting the image after edge detection is 200 × 200 middle region, by pictures all in extracted region
The gray value of element is added up to obtain intermediate pixel accumulated value；
Noise is removed using image smoothing method to collected target image, geometry pretreatment is then carried out, due to all
Object in image is transformed to unified middle position, then carries out gray processing processing, reuse by the background all same of figure
Laplacian operators carry out edge detection；
Step 3：Build the mechanical arm crawl optimal path model based on extreme learning machine；
All kinds of objects to be captured are captured using mechanical arm, crawl path sample are obtained, to capture the machinery in the sample of path
Arm starting point, final position coordinate and movement rotation angle matrix are respectively as data are output and input, with the rotation angle in each joint
The minimum object function of summation is spent, extreme learning machine is trained, it is optimal to obtain the mechanical arm crawl based on extreme learning machine
Path model；
The crawl path sample is included in the three-dimensional system of coordinate that mechanical arm tail end is built in step 1, captures starting point and grabs
Take final position coordinate and the movement rotation angle matrix in each joint of mechanical arm, the line number of the movement rotation angle matrix
It is respectively the action frequency during mechanical arm cradle head number and crawl object with columns；
Step 4：Object target image identification object classification to be captured is acquired in real time；
Object target image to be captured is acquired using the binocular ZED for being used to build 3 D stereo coordinate system in step 1, according to
The intermediate pixel accumulated value in processing procedure extraction present image in step 2, and input described based on fuzzy neural network
In object target identification model to be captured, object classification information is obtained；
Step 5；Object center of gravity is determined based on object classification, acquires crawl terminal point coordinate, mechanical arm is obtained and captures optimal road
Diameter, driving mechanical arm crawl object；
Object center of gravity is determined based on object classification information so that the crawl center counterpart weight heart of mechanical arm clamping jaw, from
And determine coordinate of the mechanical arm clamping jaw in crawl terminal, and input the mechanical arm crawl optimal path based on extreme learning machine
In model, mechanical arm crawl optimal path is obtained, and be sent to mechanical arm control system, driving mechanical arm crawl object.
Further, fuzzy neural network in the object target identification model to be captured based on fuzzy neural network
Weights, membership function mean value and variance optimize acquisition using water round-robin algorithm：
Step A1：Using rainfall layer as the weights of the fuzzy neural network, membership function mean value and variance, initialization
Rainfall layer population, and rainfall layer parameter and population is set；
The value range of rainfall layer population scale is [30,120], and the value range of river and ocean is [5,20], ocean
Number 1, minimum dminValue range for [0.025,0.1], the value range of maximum iteration is [250,500], maximum
The value range of search precision is [0.01,0.07]；
Step A2：Fitness function is set, and determines initial optimal rainfall layer and iterations t, t=1；
The corresponding weights of rainfall layer, membership function mean value and variance are substituted into the object to be captured based on fuzzy neural network
In body Model of Target Recognition, and it is defeated using the object target identification model to be captured based on fuzzy neural network that rainfall layer determines
Go out the other binary number of object type, the binary number of object classification and the difference of practical corresponding object classification binary number will be exported
The inverse of value is as the first fitness function f1(x)；
AiRepresent the i-th bit of calculating number value, BiRepresent the i-th bit of actual number value, n=6；
The fitness of each rainfall layer is calculated using the first fitness function, using the corresponding rainfall layer of maximum adaptation degree as
Sea, using the corresponding rainfall layer of secondary small fitness as river, remaining rainfall layer is as the streams for flowing into river or ocean；
Step A3：Streams is made to import river, if it find that the solution in streams is more preferable than the solution in river, then they intercourse position
It puts；
Step A4：River is made to flow into ocean, if the solution in river is more excellent than the solution of ocean, river exchanges position with ocean,
Using final ocean as optimal solution；
Step A5：It checks whether and meets evaporation conditions：Judge whether the absolute value of the difference of the adaptive value of river and ocean is small
In minimum dmin；
If it is less, thinking to meet evaporation conditions, remove the river, and re-start rainfall, random generation is new
Rainfall layer, recalculate the fitness of each rainfall layer in rainfall layer population, otherwise return to step A3, reduces dmin, into step
Rapid A6；
The new rainfall layer number generated at random is identical with the river quantity deleted；
Step A6：Judge whether to reach maximum iteration, if reaching, output global optimum sea is corresponding based on fuzzy
Weights, membership function mean value and the variance of the object target identification model to be captured of neural network, if not up to, enabling t=t
+ 1, A3 is entered step, continues next iteration.
Further, the mechanical arm based on extreme learning machine captures the power of extreme learning machine in optimal path model
Value, threshold value optimize acquisition using krill algorithm：
Step B1：Weights, threshold value using krill individual as the extreme learning machine, random initializtion krill population simultaneously set
Krill parameter and population is put, krill population includes multiple krills individual；
The value range of krill population scale is [40,300], induces inertia weight wyValue range for [0.4,0.8],
Look for food inertia weight wmValue range for [0.4,0.8], maximum induced velocity YmaxValue range for [0.02,0.07], most
The big speed M that looks for foodmaxValue range for [0.02,0.07], maximally diffuse speed DmaxValue range for [0.002,0.01],
The C of step-length zoom factortValue range is [0.1,1.5], and the value range of maximum iteration T is [300,500], and the time is normal
It measures as Δ t=0.4；
Step B2：Fitness function is set, determines initial optimal krill body position and iterations t, t=1；
The corresponding weights of krill individual and threshold value are substituted into the mechanical arm based on extreme learning machine and capture optimal path model
In, and the crawl optimal path model of the mechanical arm based on extreme learning machine determined using krill individual is obtained joint of mechanical arm and turned
Dynamic angle, using the inverse of the sum of obtained joint of mechanical arm rotational angle as the second fitness function f2(x)；
The mechanical arm articulate rotational angle of institute and smaller, the krill individual is more outstanding.
Step B3：Krill carries out induced motion, foraging activity and STOCHASTIC DIFFUSION, the position and speed to krill individual into
Row update, according to the second fitness function f2(x) current optimal krill position is determined；
Krill speed and position are changed by following three movements：
(1) krill is induced by krill around, is moved to krill around, direction αi, induced motion is carried out, formula is such as
Under：
(2) krill is carried out foraging activity, direction β by the attraction of " food "i, formula is as follows：
(3) krill carries out STOCHASTIC DIFFUSION, is randomly derived direction δi, formula is as follows：
WhereinInduced velocity during for iterations t, speed of looking for food and diffusion velocity,For
Induced velocity during iterations t+1, speed of looking for food and diffusion velocity, αi, βi, δiRespectively represent induction direction, look for food direction and
Dispersal direction, t be current iteration number, tmaxFor maximum iteration；
The movement velocity of krill is collectively formed by three movement velocity components：
Location update formula is obtained according to above formula：
Step B4：Judge whether to meet maximum iteration, if not satisfied, then t=t+1, return to step B3, until meeting
After maximum iteration, export in the crawl optimal path model of the mechanical arm based on extreme learning machine that optimal krill individual represents
The weights and threshold value of extreme learning machine.
Further, during mechanical arm captures object, if binocular ZED, which takes working space, there is barrier,
The deep image information of Use barriers object builds barrier model in the three-dimensional system of coordinate, passes through disturbance in judgement object model
Whether envelope line position is in the crawl path moving region of mechanical arm, if being not at, mechanical arm continues to capture road by original
Diameter carries out crawl work, if being in, with the envelope of barrier model and the crawl point of contact of moving region and both ends crosspoint
The intermediate point in avoidance path, the input of beginning and end coordinate are based on by the intermediate point as avoidance path, beginning and end successively
The mechanical arm avoidance path model of support vector machines obtains avoidance path；
The acquisition process of intermediate point of the envelope of the barrier model with capturing moving region is as follows：
Barrier is blocked into blocking starting point, blocking terminal and be linked to be and block straight line for mechanical arm original crawl path, is obtained all
The point tangent with the envelope of barrier model with the plane of straight line parallel is blocked, selects and blocks the nearest point of contact of straight line；
The starting point in avoidance path blocks starting point Forward 3cm to be described, and terminal is blocked to be moved after intersecting terminal in avoidance path
3cm；
The avoidance path is located at outside point of contact；
The mechanical arm avoidance path model based on support vector machines is to be located at the operation interval with all kinds of barriers
In, the envelope of barrier obtained after original mechanical arm crawl path is blocked avoidance path starting point, avoidance path termination and
With envelope with the original point of contact for capturing moving region as input data, the movement rotation in mechanical arm each joint during avoidance
Gyration matrix is minimum as object function using the sum of the rotation angle in each joint as output data, to support vector machines into
Row training obtains.
Further, in the mechanical arm avoidance path model based on support vector machines support vector machines punishment system
Number, nuclear parameter are using adaptive flower pollination algorithm optimization：
Step C1：Using pollen position as support vector machines in the mechanical arm avoidance path model based on support vector machines
Penalty coefficient, nuclear parameter initialize parameter and population；
Population number N ∈ [100,300], the initial value ε of mutagenic factor are set0∈ [0.28,0.67], transition probability value model
P ∈ [0.26,0.89] are enclosed, and it is the condition that algorithm terminates to set maximum iteration t=1500；
Step C2：Fitness function, the position of each pollen of random initializtion are set, and calculate the fitness of each pollen
Functional value, t=1；
The penalty coefficient c of the corresponding support vector machines in each pollen position, nuclear parameter g are substituted into based on supporting vector
In the mechanical arm avoidance path model of machine, the mechanical arm avoidance path model based on support vector machines of pollen location determination is utilized
The inverse of the sum of obtained joint of mechanical arm rotational angle is as third fitness function f3(x)；
Step C3：Generate random number rand1∈ [0,1] changes conversion factor, adjusts turn of global search and local search
It changes；
P=0.65+0.25*rand1
Step C4：Generate random number rand2∈ [0,1], if conversion factor p ＞ rand2, then carry out global search and perform friendship
Fork pollination operation, generates new pollen in the overall situation, otherwise carries out local search and performs self-pollination operation, is produced in local location
Raw new pollen；
Step C5：The fitness function value of each pollen is calculated, and finds out current optimal solution；
Step C6：Judge whether to meet the condition that cycle terminates, if not satisfied, C3, t=t+1 are gone to step, if satisfied, turning
Step C7；
Step C7：Export optimal location pollen, and obtain the corresponding support vector machines of optimal pollen penalty coefficient c,
Penalty coefficients of the nuclear parameter g as the support vector machines in the mechanical arm avoidance path model based on support vector machines, nuclear parameter.
Advantageous effect
The present invention provides a kind of manipulator motion planning Swarm Intelligent Computation methods, include the following steps：Step 1：Structure
The three-dimensional system of coordinate of working space where building mechanical arm；Step 2：The object target to be captured based on fuzzy neural network is built to know
Other model；Step 3：Build the mechanical arm crawl optimal path model based on extreme learning machine；Step 4：It acquires in real time and waits to capture
Object target image identifies object classification；Step 5；Object center of gravity is determined based on object classification, acquires crawl terminal point coordinate, is obtained
Mechanical arm captures optimal path, driving mechanical arm crawl object；It is had the following advantages relative to the prior art：
(1) this method can be effectively improved existing machinery arm policer operation environment, realize the intelligence manufacture of factory, replace work
People completes that classification is identified using the method for machine learning to target object；
(2) mechanical arm crawl target object one collisionless, kinetic characteristics are calculated using intelligent algorithm in this method
Meet margin requirement, path length and the shorter ideal trajectory of run duration and substantially increase production efficiency, while saved people
Member's cost, income is brought to factory；
(3) depth information in vision is obtained using the binocular zed for being capable of overlength distance perception, extends mechanical arm
Working space ensure that the job security of mechanical arm, while can provide Multi-angle working spatial information, and realizing can not in people
Ensure the normal operation of mechanical arm in the environment of, provide the possibility for adapting to complex work；
(4) complexity of Mechanical transmission test equation is solved due to establishing D-H Mo Xing, often increases one degree of freedom, equation
It solves difficulty and holds the raising of geometry difficulty, and being simple and efficient for manipulator motion formula is solved using the method for machine learning, it can
Effectively substitute traditional method for establishing D-H Mo Xing；
(5) arm for having barrier avoiding function is effectively guaranteed the safety of mechanical arm work, reduces accident generation
May, improve industrial production efficiency.
(6) for fuzzy neural network, extreme learning machine and support vector machines arrange in pairs or groups respectively optimization algorithm carry out parameter it is excellent
Change, enhance the speed of service and operation precision of the mechanical arm in actual working environment, reduce the reaction time, improve effect
Rate.
Description of the drawings
Fig. 1 is the flow chart of the method for the invention.
Specific embodiment
The present invention is described further below in conjunction with drawings and examples.
As shown in Figure 1, a kind of manipulator motion planning Swarm Intelligent Computation method, includes the following steps：
Step 1：The three-dimensional system of coordinate of working space where building mechanical arm；
Utilize a left side for the binocular ZED cameras of the entire working space of image acquisition region covering mechanical arm crawl target object
Right camera line midpoint as origin, using the right camera centers of binocular ZED to the left camera lines of centres of binocular ZED as y-axis
Forward direction establishes the 3 D stereo coordinate system of mechanical arm working space according to the right-hand rule；
Step 2：Build the object target identification model to be captured based on fuzzy neural network；
Using the intermediate pixel accumulated value of every width object target image to be captured and corresponding object classification as input
And output data, fuzzy neural network is trained, obtains object target the to be captured identification mould based on fuzzy neural network
Type；
Assuming that the mechanical arm course of work has k class target objects to need to identify, output corresponds to
Wherein, number of bits is k, i.e., corresponding fuzzy neural network has k output neuron.
The acquisition process of the intermediate pixel accumulated value of the object target image to be captured is as follows：
All kinds of object target images to be captured are acquired using the binocular ZED for being used to build 3 D stereo coordinate system in step 1
Image under various poses；It treats crawl object target image and carries out denoising, evolution, gray proces and edge inspection successively
It surveys；The size for extracting the image after edge detection is 200 × 200 middle region, by pictures all in extracted region
The gray value of element is added up to obtain intermediate pixel accumulated value；
Noise is removed using image smoothing method to collected target image, geometry pretreatment is then carried out, due to all
Object in image is transformed to unified middle position, then carries out gray processing processing, reuse by the background all same of figure
Laplacian operators carry out edge detection；
Weights of fuzzy neural network in the object target identification model to be captured based on fuzzy neural network are subordinate to
Function mean value and variance optimize acquisition using water round-robin algorithm：
Step A1：Using rainfall layer as the weights of the fuzzy neural network, membership function mean value and variance, initialization
Rainfall layer population, and rainfall layer parameter and population is set；
The value range of rainfall layer population scale is [30,120], and the value range of river and ocean is [5,20], ocean
Number 1, minimum dminValue range for [0.025,0.1], the value range of maximum iteration is [250,500], maximum
The value range of search precision is [0.01,0.07]；
Step A2：Fitness function is set, and determines initial optimal rainfall layer and iterations t, t=1；
The corresponding weights of rainfall layer, membership function mean value and variance are substituted into the object to be captured based on fuzzy neural network
In body Model of Target Recognition, and it is defeated using the object target identification model to be captured based on fuzzy neural network that rainfall layer determines
Go out the other binary number of object type, the binary number of object classification and the difference of practical corresponding object classification binary number will be exported
The inverse of value is as the first fitness function f1(x)；
AiRepresent the i-th bit of calculating number value, BiRepresent the i-th bit of actual number value, n=6；
The fitness of each rainfall layer is calculated using the first fitness function, using the corresponding rainfall layer of maximum adaptation degree as
Sea, using the corresponding rainfall layer of secondary small fitness as river, remaining rainfall layer is as the streams for flowing into river or ocean；
Step A3：Streams is made to import river, if it find that the solution in streams is more preferable than the solution in river, then they intercourse position
It puts；
Step A4：River is made to flow into ocean, if the solution in river is more excellent than the solution of ocean, river exchanges position with ocean,
Using final ocean as optimal solution；
Step A5：It checks whether and meets evaporation conditions：Judge whether the absolute value of the difference of the adaptive value of river and ocean is small
In minimum dmin；
If it is less, thinking to meet evaporation conditions, remove the river, and re-start rainfall, random generation is new
Rainfall layer, recalculate the fitness of each rainfall layer in rainfall layer population, otherwise return to step A3, reduces dmin, into step
Rapid A6；
The new rainfall layer number generated at random is identical with the river quantity deleted；
Step A6：Judge whether to reach maximum iteration, if reaching, output global optimum sea is corresponding based on fuzzy
Weights, membership function mean value and the variance of the object target identification model to be captured of neural network, if not up to, enabling t=t
+ 1, A3 is entered step, continues next iteration.
Step 3：Build the mechanical arm crawl optimal path model based on extreme learning machine；
All kinds of objects to be captured are captured using mechanical arm, crawl path sample are obtained, to capture the machinery in the sample of path
Arm starting point, final position coordinate and movement rotation angle matrix are respectively as data are output and input, with the rotation angle in each joint
The minimum object function of summation is spent, extreme learning machine is trained, it is optimal to obtain the mechanical arm crawl based on extreme learning machine
Path model；
The crawl path sample is included in the three-dimensional system of coordinate that mechanical arm tail end is built in step 1, captures starting point and grabs
Take final position coordinate and the movement rotation angle matrix in each joint of mechanical arm, the line number of the movement rotation angle matrix
It is respectively the action frequency during mechanical arm cradle head number and crawl object with columns；
The weights of extreme learning machine, threshold value use in the mechanical arm crawl optimal path model based on extreme learning machine
Krill algorithm optimizes acquisition：
Step B1：Weights, threshold value using krill individual as the extreme learning machine, random initializtion krill population simultaneously set
Krill parameter and population is put, krill population includes multiple krills individual；
The value range of krill population scale is [40,300], induces inertia weight wyValue range for [0.4,0.8],
Look for food inertia weight wmValue range for [0.4,0.8], maximum induced velocity YmaxValue range for [0.02,0.07], most
The big speed M that looks for foodmaxValue range for [0.02,0.07], maximally diffuse speed DmaxValue range for [0.002,0.01],
The C of step-length zoom factortValue range is [0.1,1.5], and the value range of maximum iteration T is [300,500], and the time is normal
It measures as Δ t=0.4；
Step B2：Fitness function is set, determines initial optimal krill body position and iterations t, t=1；
The corresponding weights of krill individual and threshold value are substituted into the mechanical arm based on extreme learning machine and capture optimal path model
In, and the crawl optimal path model of the mechanical arm based on extreme learning machine determined using krill individual is obtained joint of mechanical arm and turned
Dynamic angle, using the inverse of the sum of obtained joint of mechanical arm rotational angle as the second fitness function f2(x)；
That is the mechanical arm articulate rotational angle of institute and smaller, the krill individual is more outstanding.
Step B3：Krill carries out induced motion, foraging activity and STOCHASTIC DIFFUSION, the position and speed to krill individual into
Row update, according to the second fitness function f2(x) current optimal krill position is determined；
Krill speed and position are changed by following three movements：
(4) krill is induced by krill around, is moved to krill around, direction αi, induced motion is carried out, formula is such as
Under：
(5) krill is carried out foraging activity, direction β by the attraction of " food "i, formula is as follows：
(6) krill carries out STOCHASTIC DIFFUSION, is randomly derived direction δi, formula is as follows：
WhereinInduced velocity during for iterations t, speed of looking for food and diffusion velocity,For
Induced velocity during iterations t+1, speed of looking for food and diffusion velocity, αi, βi, δiRespectively represent induction direction, look for food direction and
Dispersal direction, t be current iteration number, tmaxFor maximum iteration；
The movement velocity of krill is collectively formed by three movement velocity components：
Location update formula is obtained according to above formula：
Step B4：Judge whether to meet maximum iteration, if not satisfied, then t=t+1, return to step B3, until meeting
After maximum iteration, export in the crawl optimal path model of the mechanical arm based on extreme learning machine that optimal krill individual represents
The weights and threshold value of extreme learning machine.
Step 4：Object target image identification object classification to be captured is acquired in real time；
Object target image to be captured is acquired using the binocular ZED for being used to build 3 D stereo coordinate system in step 1, according to
The intermediate pixel accumulated value in processing procedure extraction present image in step 2, and input described based on fuzzy neural network
In object target identification model to be captured, object classification information is obtained；
Step 5；Object center of gravity is determined based on object classification, acquires crawl terminal point coordinate, mechanical arm is obtained and captures optimal road
Diameter, driving mechanical arm crawl object；
Object center of gravity is determined based on object classification information so that the crawl center counterpart weight heart of mechanical arm clamping jaw, from
And determine coordinate of the mechanical arm clamping jaw in crawl terminal, and input the mechanical arm crawl optimal path based on extreme learning machine
In model, mechanical arm crawl optimal path is obtained, and be sent to mechanical arm control system, driving mechanical arm crawl object.
During mechanical arm captures object, if binocular ZED takes working space and barrier, Use barriers object occurs
Deep image information barrier model is built in the three-dimensional system of coordinate, pass through the envelope line position of disturbance in judgement object model
Whether it is in the crawl path moving region of mechanical arm, if being not at, mechanical arm continues to be captured by original crawl path
Work, if being in, using the envelope of barrier model with the point of contact of crawl moving region and both ends crosspoint successively as keeping away
Hinder intermediate point, the beginning and end in path, by the intermediate point in avoidance path, the input of beginning and end coordinate based on support vector machines
Mechanical arm avoidance path model, obtain avoidance path；
The acquisition process of intermediate point of the envelope of the barrier model with capturing moving region is as follows：
Barrier is blocked into blocking starting point, blocking terminal and be linked to be and block straight line for mechanical arm original crawl path, is obtained all
The point tangent with the envelope of barrier model with the plane of straight line parallel is blocked, selects and blocks the nearest point of contact of straight line；
The starting point in avoidance path blocks starting point Forward 3cm to be described, and terminal is blocked to be moved after intersecting terminal in avoidance path
3cm；
The avoidance path is located at outside point of contact；
The mechanical arm avoidance path model based on support vector machines is to be located at the operation interval with all kinds of barriers
In, the envelope of barrier obtained after original mechanical arm crawl path is blocked avoidance path starting point, avoidance path termination and
With envelope with the original point of contact for capturing moving region as input data, the movement rotation in mechanical arm each joint during avoidance
Gyration matrix is minimum as object function using the sum of the rotation angle in each joint as output data, to support vector machines into
Row training obtains.
The penalty coefficient of support vector machines, nuclear parameter are adopted in the mechanical arm avoidance path model based on support vector machines
With adaptive flower pollination algorithm optimization：
Step C1：Using pollen position as support vector machines in the mechanical arm avoidance path model based on support vector machines
Penalty coefficient, nuclear parameter initialize parameter and population；
Population number N ∈ [100,300], the initial value ε of mutagenic factor are set0∈ [0.28,0.67], transition probability value model
P ∈ [0.26,0.89] are enclosed, and it is the condition that algorithm terminates to set maximum iteration t=1500；
Step C2：Fitness function, the position of each pollen of random initializtion are set, and calculate the fitness of each pollen
Functional value, t=1；
The penalty coefficient c of the corresponding support vector machines in each pollen position, nuclear parameter g are substituted into based on supporting vector
In the mechanical arm avoidance path model of machine, the mechanical arm avoidance path model based on support vector machines of pollen location determination is utilized
The inverse of the sum of obtained joint of mechanical arm rotational angle is as third fitness function f3(x)；
That is the articulate rotational angle of institute of mechanical arm avoidance and smaller, the pollen individual is more outstanding；
Step C3：Generate random number rand1∈ [0,1] changes conversion factor, adjusts turn of global search and local search
It changes；
P=0.65+0.25*rand1
Step C4：Generate random number rand2∈ [0,1], if conversion factor p ＞ rand2, then carry out global search and perform friendship
Fork pollination operation, generates new pollen in the overall situation, otherwise carries out local search and performs self-pollination operation, is produced in local location
Raw new pollen；
Step C5：The fitness function value of each pollen is calculated, and finds out current optimal solution；
Step C6：Judge whether to meet the condition that cycle terminates, if not satisfied, C3, t=t+1 are gone to step, if satisfied, turning
Step C7；
Step C7：Export optimal location pollen, and obtain the corresponding support vector machines of optimal pollen penalty coefficient c,
Penalty coefficients of the nuclear parameter g as the support vector machines in the mechanical arm avoidance path model based on support vector machines, nuclear parameter.
The present invention is described in detail above in association with specific embodiment, these not form the limitation to invention.
Without departing from the principles of the present invention, those skilled in the art can also make many modification and improvement, these also should
It belongs to the scope of protection of the present invention.
Claims (5)
- A kind of 1. manipulator motion planning Swarm Intelligent Computation method, which is characterized in that include the following steps：Step 1：The three-dimensional system of coordinate of working space where building mechanical arm；It is taken the photograph using the left and right of the binocular ZED cameras of the entire working space of image acquisition region covering mechanical arm crawl target object It is positive by y-axis of the right camera centers of binocular ZED to the left camera lines of centres of binocular ZED as head line midpoint is as origin, The 3 D stereo coordinate system of mechanical arm working space is established according to the right-hand rule；Step 2：Build the object target identification model to be captured based on fuzzy neural network；Using the intermediate pixel accumulated value of every width object target image to be captured and corresponding object classification as input and defeated Go out data, fuzzy neural network is trained, obtain the object target identification model to be captured based on fuzzy neural network；The acquisition process of the intermediate pixel accumulated value of the object target image to be captured is as follows：All kinds of object target images to be captured are acquired each using the binocular ZED for being used to build 3 D stereo coordinate system in step 1 Image under kind pose；It treats crawl object target image and carries out denoising, evolution, gray proces and edge detection successively； The size for extracting the image after edge detection is 200 × 200 middle region, by all pixels in extracted region Gray value is added up to obtain intermediate pixel accumulated value；Step 3：Build the mechanical arm crawl optimal path model based on extreme learning machine；All kinds of objects to be captured are captured using mechanical arm, obtain crawl path sample, are risen with capturing the mechanical arm in the sample of path Point, final position coordinate and movement rotation angle matrix are total with the rotation angle in each joint respectively as outputting and inputting data With minimum object function, extreme learning machine is trained, obtains the mechanical arm crawl optimal path based on extreme learning machine Model；The crawl path sample is included in the three-dimensional system of coordinate that mechanical arm tail end is built in step 1, and crawl starting point and crawl are eventually Point position coordinates and the movement rotation angle matrix in each joint of mechanical arm, the line number and row of the movement rotation angle matrix Number is respectively mechanical arm cradle head number and captures the action frequency during object；Step 4：Object target image identification object classification to be captured is acquired in real time；Object target image to be captured is acquired using the binocular ZED for being used to build 3 D stereo coordinate system in step 1, according to step The intermediate pixel accumulated value in processing procedure extraction present image in 2, and wait to grab based on fuzzy neural network described in input It takes in object target identification model, obtains object classification information；Step 5；Object center of gravity is determined based on object classification, acquires crawl terminal point coordinate, mechanical arm crawl optimal path is obtained, drives Dynamic mechanical arm crawl object；Object center of gravity is determined based on object classification information so that the crawl center counterpart weight heart of mechanical arm clamping jaw, so as to really Determine coordinate of the mechanical arm clamping jaw in crawl terminal, and input the mechanical arm crawl optimal path model based on extreme learning machine In, mechanical arm crawl optimal path is obtained, and be sent to mechanical arm control system, driving mechanical arm crawl object.
- 2. the according to the method described in claim 1, it is characterized in that, object target to be captured based on fuzzy neural network The weights of fuzzy neural network, membership function mean value and variance optimize acquisition using water round-robin algorithm in identification model：Step A1：Using rainfall layer as the weights of the fuzzy neural network, membership function mean value and variance, rainfall is initialized Layer population, and rainfall layer parameter and population is set；The value range of rainfall layer population scale is [30,120], and the value range of river and ocean is [5,20], ocean number 1, minimum dminValue range for [0.025,0.1], the value range of maximum iteration is [250,500], maximum search The value range of precision is [0.01,0.07]；Step A2：Fitness function is set, and determines initial optimal rainfall layer and iterations t, t=1；The corresponding weights of rainfall layer, membership function mean value and variance are substituted into the object mesh to be captured based on fuzzy neural network It marks in identification model, and the object target identification model output to be captured based on fuzzy neural network determined using rainfall layer The binary number of body classification will export the binary number of object classification and the difference of practical corresponding object classification binary number Inverse is used as the first fitness function f1(x)；AiRepresent the i-th bit of calculating number value, BiRepresent the i-th bit of actual number value, n=6；The fitness of each rainfall layer is calculated using the first fitness function, using the corresponding rainfall layer of maximum adaptation degree as greatly Sea, using the corresponding rainfall layer of secondary small fitness as river, remaining rainfall layer is as the streams for flowing into river or ocean；Step A3：Streams is made to import river, if it find that the solution in streams is more preferable than the solution in river, then they intercourse position；Step A4：River is made to flow into ocean, if the solution in river is more excellent than the solution of ocean, river exchanges position with ocean, with most Whole ocean is as optimal solution；Step A5：It checks whether and meets evaporation conditions：Judge whether the absolute value of the difference of the adaptive value of river and ocean is less than pole Small value dmin；If it is less, thinking to meet evaporation conditions, remove the river, and re-start rainfall, generate new drop at random Rain layer, recalculates the fitness of each rainfall layer in rainfall layer population, otherwise return to step A3, reduces dmin, enter step A6；The new rainfall layer number generated at random is identical with the river quantity deleted；Step A6：Judge whether to reach maximum iteration, if reaching, output global optimum sea is corresponding to be based on fuzzy neural Weights, membership function mean value and the variance of the object target identification model to be captured of network, if not up to, enabling t=t+1, A3 is entered step, continues next iteration.
- 3. according to the method described in claim 1, it is characterized in that, the mechanical arm based on extreme learning machine captures optimal road The weights of extreme learning machine, threshold value optimize acquisition using krill algorithm in diameter model：Step B1：Weights, threshold value using krill individual as the extreme learning machine, random initializtion krill population simultaneously set phosphorus Shrimp species swarm parameter, krill population include multiple krills individual；The value range of krill population scale is [40,300], induces inertia weight wyValue range for [0.4,0.8], look for food Inertia weight wmValue range for [0.4,0.8], maximum induced velocity YmaxValue range for [0.02,0.07], maximum is looked for Eat speed MmaxValue range for [0.02,0.07], maximally diffuse speed DmaxValue range for [0.002,0.01], step-length The C of zoom factortValue range is [0.1,1.5], and the value range of maximum iteration T is [300,500], and time constant is Δ t=0.4；Step B2：Fitness function is set, determines initial optimal krill body position and iterations t, t=1；The corresponding weights of krill individual and threshold value are substituted into the mechanical arm crawl optimal path model based on extreme learning machine, and The crawl optimal path model of the mechanical arm based on extreme learning machine determined using krill individual obtains joint of mechanical arm angle of rotation Degree, using the inverse of the sum of obtained joint of mechanical arm rotational angle as the second fitness function f2(x)；Step B3：Krill carries out induced motion, foraging activity and STOCHASTIC DIFFUSION, and the position and speed to krill individual carry out more Newly, according to the second fitness function f2(x) current optimal krill position is determined；Step B4：Judge whether to meet maximum iteration, if not satisfied, then t=t+1, return to step B3, maximum until meeting After iterations, the limit in the crawl optimal path model of the mechanical arm based on extreme learning machine that optimal krill individual represents is exported The weights and threshold value of learning machine.
- 4. according to claim 1-3 any one of them methods, which is characterized in that during mechanical arm captures object, if double Mesh ZED takes working space and barrier occurs, then the deep image information of Use barriers object structure in the three-dimensional system of coordinate Barrier model is built, whether the crawl path moving region of mechanical arm is in by the envelope line position of disturbance in judgement object model In, if being not at, mechanical arm continues to carry out crawl work by original crawl path, if being in, with the envelope of barrier model Line and the point of contact of crawl moving region and both ends the crosspoint intermediate point as avoidance path, beginning and end successively, by avoidance Intermediate point, the mechanical arm avoidance path model of the beginning and end coordinate input based on support vector machines in path, obtain avoidance road Diameter；The acquisition process of intermediate point of the envelope of the barrier model with capturing moving region is as follows：Barrier is blocked into blocking starting point, blocking terminal and be linked to be and block straight line for mechanical arm original crawl path, obtains all and cuts The plane of the disconnected straight line parallel point tangent with the envelope of barrier model, selects and blocks the nearest point of contact of straight line；The starting point in avoidance path blocks starting point Forward 3cm to be described, and terminal is blocked to move 3cm after intersecting terminal in avoidance path；The avoidance path is located at outside point of contact；The mechanical arm avoidance path model based on support vector machines is located in the operation interval with all kinds of barriers, The envelope of barrier by original mechanical arm crawl path block after obtain avoidance path starting point, avoidance path termination and with packet The point of contact of winding thread and original crawl moving region is as input data, the movement rotation angle in mechanical arm each joint during avoidance Matrix is spent as output data, and using the sum of the rotation angle in each joint minimum as object function, support vector machines is instructed Practice and obtain.
- 5. the according to the method described in claim 4, it is characterized in that, mechanical arm avoidance path mould based on support vector machines The penalty coefficient of support vector machines, nuclear parameter are using adaptive flower pollination algorithm optimization in type：Step C1：Using pollen position as the punishment of support vector machines in the mechanical arm avoidance path model based on support vector machines Coefficient, nuclear parameter initialize parameter and population；Population number N ∈ [100,300], the initial value ε of mutagenic factor are set0∈ [0.28,0.67], transition probability value range p ∈ [0.26,0.89], and it is the condition that algorithm terminates to set maximum iteration t=1500；Step C2：Fitness function, the position of each pollen of random initializtion are set, and calculate the fitness function of each pollen Value, t=1；The penalty coefficient c of the corresponding support vector machines in each pollen position, nuclear parameter g are substituted into based on support vector machines In mechanical arm avoidance path model, obtained using the mechanical arm avoidance path model based on support vector machines of pollen location determination The sum of joint of mechanical arm rotational angle inverse as third fitness function f3(x)；Step C3：Generate random number rand1∈ [0,1] changes conversion factor, adjusts global search and the conversion of local search；P=0.65+0.25*rand1Step C4：Generate random number rand2∈ [0,1], if conversion factor p ＞ rand2, then carry out global search execution intersection and award Powder operates, and new pollen is generated in the overall situation, otherwise carries out local search and performs self-pollination operation, is generated in local location new Pollen；Step C5：The fitness function value of each pollen is calculated, and finds out current optimal solution；Step C6：Judge whether to meet the condition that cycle terminates, if not satisfied, C3, t=t+1 are gone to step, if satisfied, going to step C7；Step C7：Optimal location pollen is exported, and obtains the penalty coefficient c of the corresponding support vector machines of optimal pollen, core ginseng Penalty coefficients of the number g as the support vector machines in the mechanical arm avoidance path model based on support vector machines, nuclear parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810068243.7A CN108196453B (en) | 2018-01-24 | 2018-01-24 | Intelligent calculation method for mechanical arm motion planning group |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810068243.7A CN108196453B (en) | 2018-01-24 | 2018-01-24 | Intelligent calculation method for mechanical arm motion planning group |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108196453A true CN108196453A (en) | 2018-06-22 |
| CN108196453B CN108196453B (en) | 2020-11-20 |
Family
ID=62590540
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810068243.7A Active CN108196453B (en) | 2018-01-24 | 2018-01-24 | Intelligent calculation method for mechanical arm motion planning group |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108196453B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108830248A (en) * | 2018-06-25 | 2018-11-16 | 中南大学 | A kind of pedestrian's local feature big data mixing extracting method |
| CN109108978A (en) * | 2018-09-11 | 2019-01-01 | 武汉科技大学 | Three-dimensional space manipulator motion planning method based on study Generalization Mechanism |
| CN109693239A (en) * | 2018-12-29 | 2019-04-30 | 深圳市越疆科技有限公司 | A kind of robot grasping means based on deeply study |
| CN111504328A (en) * | 2020-05-22 | 2020-08-07 | 梅卡曼德（北京）机器人科技有限公司 | Robot motion planning method, path planning method, grabbing method and device |
| CN112614171A (en) * | 2020-11-26 | 2021-04-06 | 厦门大学 | Air-ground integrated dynamic environment sensing system for cluster operation of engineering machinery |
| CN112847358A (en) * | 2020-12-31 | 2021-05-28 | 深圳辰视智能科技有限公司 | Path planning method and industrial robot |
| CN113043286A (en) * | 2020-12-25 | 2021-06-29 | 杭州电子科技大学 | Multi-degree-of-freedom mechanical arm real-time obstacle avoidance path planning system and method |
| CN113538459A (en) * | 2021-07-07 | 2021-10-22 | 重庆大学 | Multi-mode grabbing obstacle avoidance detection optimization method based on drop point area detection |
| CN114012727A (en) * | 2021-11-11 | 2022-02-08 | 江苏昱博自动化设备有限公司 | Mechanical arm motion planning method and system |
| CN116125803A (en) * | 2022-12-28 | 2023-05-16 | 淮阴工学院 | Inverter backstepping fuzzy neural network control strategy based on extreme learning machine |
| CN117245672A (en) * | 2023-11-20 | 2023-12-19 | 南昌工控机器人有限公司 | Intelligent motion control system and method for modularized assembly of camera support |
| CN117290980A (en) * | 2023-11-27 | 2023-12-26 | 江西格如灵科技股份有限公司 | Mechanical arm simulation method and system based on Unity platform |
| CN116125803B (en) * | 2022-12-28 | 2024-06-11 | 淮阴工学院 | Inverter backstepping fuzzy neural network control method based on extreme learning machine |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102008019635A1 (en) * | 2008-04-18 | 2009-10-22 | Siemens Aktiengesellschaft | Medical workplace |
| KR101105325B1 (en) * | 2009-09-08 | 2012-01-16 | 부산대학교 산학협력단 | Method for Path-planning for Actual Robots |
| CN106874914A (en) * | 2017-01-12 | 2017-06-20 | 华南理工大学 | A kind of industrial machinery arm visual spatial attention method based on depth convolutional neural networks |
| CN106945059A (en) * | 2017-03-27 | 2017-07-14 | 中国地质大学（武汉） | A kind of gesture tracking method based on population random disorder multi-objective genetic algorithm |
| CN107169993A (en) * | 2017-05-12 | 2017-09-15 | 甘肃政法学院 | Detection recognition method is carried out to object using public security video monitoring blurred picture |
| CN107414830A (en) * | 2017-07-31 | 2017-12-01 | 中南大学 | A kind of carrying machine human arm manipulation multi-level mapping intelligent control method and system |
-
2018
- 2018-01-24 CN CN201810068243.7A patent/CN108196453B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102008019635A1 (en) * | 2008-04-18 | 2009-10-22 | Siemens Aktiengesellschaft | Medical workplace |
| KR101105325B1 (en) * | 2009-09-08 | 2012-01-16 | 부산대학교 산학협력단 | Method for Path-planning for Actual Robots |
| CN106874914A (en) * | 2017-01-12 | 2017-06-20 | 华南理工大学 | A kind of industrial machinery arm visual spatial attention method based on depth convolutional neural networks |
| CN106945059A (en) * | 2017-03-27 | 2017-07-14 | 中国地质大学（武汉） | A kind of gesture tracking method based on population random disorder multi-objective genetic algorithm |
| CN107169993A (en) * | 2017-05-12 | 2017-09-15 | 甘肃政法学院 | Detection recognition method is carried out to object using public security video monitoring blurred picture |
| CN107414830A (en) * | 2017-07-31 | 2017-12-01 | 中南大学 | A kind of carrying machine human arm manipulation multi-level mapping intelligent control method and system |
Non-Patent Citations (5)
| Title |
|---|
| 刘晴: "基于区域特征的目标跟踪算法研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
| 吴江: "移动机械臂运动规划研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 张书真 宋海龙 向晓燕 赵英男: "采用快速SIFT算法实现目标识别", 《计算机***应用》 * |
| 鄢诚挚: "基于视觉的6R工业机械臂运动规划", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 鞠初旭: "模糊神经网络的研究及应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108830248A (en) * | 2018-06-25 | 2018-11-16 | 中南大学 | A kind of pedestrian's local feature big data mixing extracting method |
| CN109108978A (en) * | 2018-09-11 | 2019-01-01 | 武汉科技大学 | Three-dimensional space manipulator motion planning method based on study Generalization Mechanism |
| CN109693239A (en) * | 2018-12-29 | 2019-04-30 | 深圳市越疆科技有限公司 | A kind of robot grasping means based on deeply study |
| WO2021232669A1 (en) * | 2020-05-22 | 2021-11-25 | 梅卡曼德（北京）机器人科技有限公司 | Robot motion planning method, path planning method, grabbing method and devices thereof |
| CN111504328A (en) * | 2020-05-22 | 2020-08-07 | 梅卡曼德（北京）机器人科技有限公司 | Robot motion planning method, path planning method, grabbing method and device |
| CN114061580B (en) * | 2020-05-22 | 2023-12-29 | 梅卡曼德（北京）机器人科技有限公司 | Robot grabbing method and device based on symmetry degree, electronic equipment and medium |
| CN114061580A (en) * | 2020-05-22 | 2022-02-18 | 梅卡曼德（北京）机器人科技有限公司 | Robot grabbing method and device based on symmetry degree, electronic equipment and medium |
| CN112614171B (en) * | 2020-11-26 | 2023-12-19 | 厦门大学 | Air-ground integrated dynamic environment sensing system for engineering machinery cluster operation |
| CN112614171A (en) * | 2020-11-26 | 2021-04-06 | 厦门大学 | Air-ground integrated dynamic environment sensing system for cluster operation of engineering machinery |
| CN113043286B (en) * | 2020-12-25 | 2022-05-03 | 杭州电子科技大学 | Multi-degree-of-freedom mechanical arm real-time obstacle avoidance path planning system and method |
| CN113043286A (en) * | 2020-12-25 | 2021-06-29 | 杭州电子科技大学 | Multi-degree-of-freedom mechanical arm real-time obstacle avoidance path planning system and method |
| CN112847358A (en) * | 2020-12-31 | 2021-05-28 | 深圳辰视智能科技有限公司 | Path planning method and industrial robot |
| CN113538459A (en) * | 2021-07-07 | 2021-10-22 | 重庆大学 | Multi-mode grabbing obstacle avoidance detection optimization method based on drop point area detection |
| CN113538459B (en) * | 2021-07-07 | 2023-08-11 | 重庆大学 | Multimode grabbing obstacle avoidance detection optimization method based on drop point area detection |
| CN114012727A (en) * | 2021-11-11 | 2022-02-08 | 江苏昱博自动化设备有限公司 | Mechanical arm motion planning method and system |
| CN116125803A (en) * | 2022-12-28 | 2023-05-16 | 淮阴工学院 | Inverter backstepping fuzzy neural network control strategy based on extreme learning machine |
| CN116125803B (en) * | 2022-12-28 | 2024-06-11 | 淮阴工学院 | Inverter backstepping fuzzy neural network control method based on extreme learning machine |
| CN117245672A (en) * | 2023-11-20 | 2023-12-19 | 南昌工控机器人有限公司 | Intelligent motion control system and method for modularized assembly of camera support |
| CN117245672B (en) * | 2023-11-20 | 2024-02-02 | 南昌工控机器人有限公司 | Intelligent motion control system and method for modularized assembly of camera support |
| CN117290980A (en) * | 2023-11-27 | 2023-12-26 | 江西格如灵科技股份有限公司 | Mechanical arm simulation method and system based on Unity platform |
| CN117290980B (en) * | 2023-11-27 | 2024-02-02 | 江西格如灵科技股份有限公司 | Mechanical arm simulation method and system based on Unity platform |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108196453B (en) | 2020-11-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108196453A (en) | A kind of manipulator motion planning Swarm Intelligent Computation method | |
| CN108247637A (en) | A kind of industrial machine human arm vision anticollision control method | |
| CN108280856B (en) | Unknown object grabbing pose estimation method based on mixed information input network model | |
| CN111618847B (en) | Mechanical arm autonomous grabbing method based on deep reinforcement learning and dynamic motion elements | |
| CN112297013B (en) | Robot intelligent grabbing method based on digital twin and deep neural network | |
| CN109483573A (en) | Machine learning device, robot system and machine learning method | |
| CN107972026B (en) | Robot, mechanical arm and control method and device thereof | |
| CN109986560B (en) | Mechanical arm self-adaptive grabbing method for multiple target types | |
| CN110298886B (en) | Dexterous hand grabbing planning method based on four-stage convolutional neural network | |
| CN110450153A (en) | A kind of mechanical arm article active pick-up method based on deeply study | |
| CN106737673A (en) | A kind of method of the control of mechanical arm end to end based on deep learning | |
| Sarantopoulos et al. | Split deep q-learning for robust object singulation | |
| CN108362284A (en) | A kind of air navigation aid based on bionical hippocampus cognitive map | |
| CN110238855A (en) | A kind of robot random ordering workpiece grabbing method based on the reverse intensified learning of depth | |
| Tang et al. | Learning collaborative pushing and grasping policies in dense clutter | |
| CN110969660A (en) | Robot feeding system based on three-dimensional stereoscopic vision and point cloud depth learning | |
| CN108229678A (en) | Network training method, method of controlling operation thereof, device, storage medium and equipment | |
| Iqbal et al. | Toward sim-to-real directional semantic grasping | |
| Ichiwara et al. | Spatial attention point network for deep-learning-based robust autonomous robot motion generation | |
| Chen et al. | Towards generalization and data efficient learning of deep robotic grasping | |
| CN111814823A (en) | Transfer learning method based on scene template generation | |
| CN113664828A (en) | Robot grabbing-throwing method based on deep reinforcement learning | |
| Ren et al. | Fast-learning grasping and pre-grasping via clutter quantization and Q-map masking | |
| CN116852347A (en) | State estimation and decision control method for non-cooperative target autonomous grabbing | |
| Wong et al. | Ant Colony Optimization and image model-based robot manipulator system for pick-and-place tasks |
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 |