CN106886155A - A kind of quadruped robot control method of motion trace based on PSO PD neutral nets - Google Patents
A kind of quadruped robot control method of motion trace based on PSO PD neutral nets Download PDFInfo
- Publication number
- CN106886155A CN106886155A CN201710295939.9A CN201710295939A CN106886155A CN 106886155 A CN106886155 A CN 106886155A CN 201710295939 A CN201710295939 A CN 201710295939A CN 106886155 A CN106886155 A CN 106886155A
- Authority
- CN
- China
- Prior art keywords
- quadruped robot
- formula
- trunk
- control
- robot
- 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
- 230000033001 locomotion Effects 0.000 title claims abstract description 105
- 230000007935 neutral effect Effects 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 32
- 230000005484 gravity Effects 0.000 claims abstract description 59
- 238000006073 displacement reaction Methods 0.000 claims abstract description 23
- 230000009466 transformation Effects 0.000 claims abstract description 8
- 239000002245 particle Substances 0.000 claims description 80
- 210000002414 leg Anatomy 0.000 claims description 41
- 238000003062 neural network model Methods 0.000 claims description 35
- 230000006870 function Effects 0.000 claims description 34
- 238000005457 optimization Methods 0.000 claims description 29
- 210000004394 hip joint Anatomy 0.000 claims description 28
- 238000004422 calculation algorithm Methods 0.000 claims description 20
- 239000011159 matrix material Substances 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 12
- 230000003044 adaptive effect Effects 0.000 claims description 10
- 230000008859 change Effects 0.000 claims description 9
- 210000000629 knee joint Anatomy 0.000 claims description 9
- 230000005021 gait Effects 0.000 claims description 7
- 210000000689 upper leg Anatomy 0.000 claims description 6
- 210000004027 cell Anatomy 0.000 claims description 5
- 238000012804 iterative process Methods 0.000 claims description 5
- 210000002569 neuron Anatomy 0.000 claims description 5
- 238000005265 energy consumption Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 4
- 230000001133 acceleration Effects 0.000 claims description 3
- 230000008929 regeneration Effects 0.000 claims description 3
- 238000011069 regeneration method Methods 0.000 claims description 3
- 230000008054 signal transmission Effects 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000011217 control strategy Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000002068 genetic effect Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000006641 stabilisation Effects 0.000 description 2
- 238000011105 stabilization Methods 0.000 description 2
- 108010023321 Factor VII Proteins 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
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)
- Manipulator (AREA)
- Feedback Control In General (AREA)
Abstract
The present invention relates to a kind of quadruped robot control method of motion trace based on PSO PD neutral nets, including:(1) target drop point of the robot on movement locus is asked for;(2) the trunk position of centre of gravity of target drop point, feedback is input into the input layer of PSO PD neutral nets;After linear transformation enter the first hidden layer, scale operation with differentiate;Enter the second hidden layer after linear transformation, obtain instructing displacement on x directions and y directions;Into the 3rd hidden layer, try to achieve and instruct shift length, instruct trunk direction;Into layer 5, to sufficient end TRAJECTORY CONTROL, the course changing control of quadruped robot;Into layer 6, quadruped robot attitude is adjusted when disturbing;Into layer 7, trunk direction, the trunk position of centre of gravity of quadruped robot are asked for, towards layer 5 is fed back to, trunk position of centre of gravity feeds back to input layer to trunk.With more preferable Nonlinear Decoupling control ability, control is accurate, and stability is strong, with preferable antijamming capability.
Description
Technical field
The present invention relates to a kind of robot motion's method for controlling trajectory, and in particular to a kind of based on PSO-PD neutral nets
Quadruped robot control method of motion trace.
Background technology
With developing rapidly for science and technology, robot technology is also increasingly being applied among complex environment.Wherein
Legged type robot is moved with certain superiority relative to wheeled or caterpillar type robot in non-structure environment.And four-footed machine
Device people with simple structure, the advantage such as motion stabilization, therefore anticipates compared to other Multifeet walking robots with research higher
Justice and application value.When complex environment works, robot runs along pre-set movement locus.Can be with most fast
Speed, minimum energy consumption passes through each operating point, and at the same time can also avoid various obstacles etc. influences stable factor.Cause
Be applied to research of the movement locus control of this research quadruped robot to quadruped robot is significant.
Quadruped robot is a nonlinear system for complexity, and trunk direction has coupled relation with position of centre of gravity, because
The control of this quadruped robot movement locus is a sufficiently complex Nonlinear Decoupling control problem.Existing method can only be to four-footed machine
Device people realizes Linear track control, or circular trace control;Or on this basis by the gait simple non-linear rail of switching realization
The control of mark.But continuous accurately nonlinear motion TRAJECTORY CONTROL cannot be realized to robot all the time.
General existing PD neutral nets are four layers of feedforward neural network, and ground floor is input layer, receive control signal with it is anti-
Feedback signal;The second layer is proportion differential computation layer, and ratio calculating and differential calculation are carried out to error by action function;Third layer
It is control rate output layer, the result differentiated by comparative example is carried out linear operation and obtains control rate;4th layer right for control
As.
The content of the invention
In view of the shortcomings of the prior art, the invention provides a kind of quadruped robot motion based on PSO-PD neutral nets
Method for controlling trajectory.
A kind of special PD neutral nets towards the control of quadruped robot movement locus are devised in the present invention, but due to this
Existence feedback cannot realize god with the action function that can discontinuously lead by BP learning algorithms in PD neural network networks
It is then of the invention to use the learning algorithm for improving PSO algorithms as PD neutral nets through the adaptive learning of network.Improve PSO
Algorithmic statement is fast, high precision and should not be absorbed in local optimum, can accurately realize the adaptive learning of PD neutral nets.Always
It, the present invention can realize that quadruped robot movement locus is controlled fast accurate, be expected to be obtained in motion planning and robot control
Extensive use.
The technical scheme is that:
A kind of quadruped robot control method of motion trace based on PSO-PD neutral nets, including:
(1) according to quadruped robot movement locus and default step-length, quadruped robot center of gravity is asked on the movement locus
Target drop point, i.e. coordinate of the quadruped robot center of gravity on the movement locus;It is for actual result.
(2) target drop point, the body of Real-time Feedback of the quadruped robot center of gravity for asking for step (1) on the movement locus
Dry position of centre of gravity (yx,yy) it is input into the input layer of PD neural network models;Into the first of PD neural network models after linear transformation
Hidden layer, carries out scale operation and differentiates in the first hidden layer;Again into PD neural network models after linear transformation
Second hidden layer, the displacement of instructing on x directions and y directions is obtained in the second hidden layer, and residing coordinate system is that quadruped robot is transported
The coordinate system of the horizontal plane where dynamic, origin is quadruped robot original center of gravity position;Into the of PD neural network models
Three hidden layers, try to achieve in the 3rd hidden layer and instruct shift length dr, instruct steering angle d θ, and it refers to four-footed to instruct shift length dr
The distance that robot is stepped from current this step to next step, instruct steering angle refer to quadruped robot from current this step to
The trunk steering angle of next step;Into the layer 5 of PD neural network models, the foot to quadruped robot is realized in layer 5
End TRAJECTORY CONTROL, course changing control;Into the layer 6 of PD neural network models, when quadruped robot foot end is slided, adjust
Whole quadruped robot attitude, it is ensured that quadruped robot is stable;But the process can make robot center of gravity change, and can be considered
The random disturbance of size that random times occur, in addition with the disturbance that many other reasonses are caused, adds in layer 6 and disturbs
It is simulated, and the step-length after disturbance will be added to feed back to layer 5 as actual step size;Into the of PD neural network models
Seven layers, ask for the trunk of quadruped robot towards yθ, trunk position of centre of gravity (yx,yy), trunk is towards yθIt refer to current end of taking a step
Afterwards, the angle of the actual direction of trunk, i.e. trunk and X-axis;Trunk position of centre of gravity (yx,yy) refer to quadruped robot center of gravity in institute
State the coordinate in coordinate system;The trunk of quadruped robot is towards yθFeed back to the layer 5 of PD neural network models, trunk center of gravity
Position (yx,yy) feed back to the input layer of PD neural network models.
According to currently preferred, using multilayer nerve in improved particle swarm optimization algorithm adjustment PD neural network models
Connection weight between unit, realizes the adaptive learning of the PD neutral nets of object manipulator movement locus control, including:
A, the On The Choice of the connection weight between multilayer neuron is converted into optimization problem, the mesh of optimization problem
Scalar functions are the L2 norms of output vector and tutor's signal vector, and as shown in formula (I), tutor's signal is target landing point coordinates, defeated
Outgoing vector is quadruped robot actual center gravity position;
In formula (I), Error is the object function of optimization problem, and xd (k), yd (k) are respectively quadruped robot kth step
Center of gravity target drop point abscissa, ordinate, yx(k)、yyK () is the reality of the center of gravity target drop point of quadruped robot kth step
Abscissa, ordinate;
B, the span that each connection weight is determined using previous experience, that is, determine Search Range;
C, random initializtion a group particle, i.e. population in Search Range, including initialization particle initial position with
Initial velocity, with these three index expression particle characteristicses of position, speed and fitness, institute in positional representation PD neural network models
Some connection weight values, speed represents the direction of each particle evolution, and fitness value is tried to achieve by fitness function, i.e. each grain
The corresponding object function of son;And each particle represents a potential optimal solution of above-mentioned optimization problem;
The speed of particle is according to the current location of particle, present speed, the history optimum position Pbest of particle and population
The position Gbest of middle optimal particle updates, the speed of particleMore new formula such as formula (II) shown in:
In formula (II), id is the numbering of particle in population,It is the speed of the i-th generation particle,For the i-th generation particle exists
I-th instead of preceding history optimum position,It is the position of optimal particle in the i-th generation population;ω (i) is the used of the i-th generation particle
Property weight, its size determines that speed to what extent inherits the movement velocity of previous generation particles;%1, %2It is acceleration factor,
Value is nonnegative constant;r1、r2It is the random number between 0 to 1;It is the position of the i-th generation particle;The value of ω during initialization
ωstartIt is 0.9, the value ω of ω at the end of iteration34dIt is 0.01, inertia weight ω accelerates decay in an iterative process, the initial stage is excellent
First speed of searching optimization, the later stage focuses on low optimization accuracy, shown in the more new formula such as formula (III) of inertia weight ω:
In formula (III), maxg;N is maximum iteration;
Present invention adds decreases in non-linear inertia weight ω (i), inertia weight ω (i) determines speed to what extent
The movement velocity of previous generation is inherited, the bigger speed of searching optimization of ω (i) is faster in an iterative process, but precision is lower;Otherwise then precision
Higher, speed is lower.To reach the balance of speed and precision, inertia weight ω (i) is larger at the iteration initial stage, as iteration is carried out
Inertia weight progressively reduces.
The speed of the particle after being updatedAfterwards, the position of the particle, the position of particle are updatedRenewal it is public
Shown in formula such as formula (IV):
If formula (IV) ask forCorresponding object function is less thanCorresponding fitness function, then
Conversely,Meanwhile, the position of optimal particle, obtains in Population Regeneration
In this way, carrying out successive ignition obtains approximate optimal solution, even if PD neural network models control error minimum is near
Like best initial weights.
After adding the thought of the TSP question in genetic algorithm, i.e. particle position to update, there is certain probability to occur
Variation in span.TSP question has expanded in iteration the population search space for constantly reducing, can be
Bigger space development search, makes it jump out local optimum, and increase obtains the possibility of global optimum.Each particle in position more
After new, a random number can be generated, if greater than 0.9 random initializtion particle again in Search Range.If less than
Doed nothing equal to 0.9.
Improved particle swarm optimization algorithm takes into account convergence rate and low optimization accuracy while being difficult to be absorbed in local optimum.With
It substitutes the adaptive learning that traditional PS O particle cluster algorithms realize PSO-PD neutral nets.Improved particle swarm optimization algorithm, learns
The habit time is shorter, and control accuracy is higher, therefore can better meet the demand for control of quadruped robot movement locus control.
According to currently preferred, the step (1), according to quadruped robot movement locus, ask for quadruped robot and exist
Target drop point on the movement locus, the movement locus for setting quadruped robot is sine curve, during along sinusoidal motion, machine
The radius of turn of device people, angle of turn can be used to prove whether the control strategy is generally applicable all constantly changing
In most nonlinear motion tracks.The span for setting default step-length L is 0.38-0.42m, when no external disturbance
When, the center of gravity of robot will not change;Ask for the formula such as formula of target drop point of the quadruped robot on the movement locus
(V) shown in:
In formula (V), x_1, y_1 are respectively the abscissa and ordinate of back quadruped robot center of gravity, and x, y are respectively
When the abscissa and ordinate of back quadruped robot center of gravity;
The initial value of x_1, y_1 is 0, after asking for x, y, then is assigned to x_1, y_1, iterates after 200 times, obtains
Target drop point of the quadruped robot on the movement locus.
It is further preferred that the value of L is 0.4m.
According to currently preferred, tried to achieve in the 3rd hidden layer and instruct shift length dr, instruct trunk towards d θ, including:
Ask for instructing shift length dr by formula (VI), formula (VI) is as follows:
In formula (VI), d Δs x, d Δ y is instructed x increments and is instructed y increments, i.e. x directions and y by what the second hidden layer was tried to achieve
Displacement is instructed on direction;
Ask for instructing trunk towards d θ by formula (VII), formula (VII) is as follows:
According to currently preferred, realize the sufficient end TRAJECTORY CONTROL to quadruped robot, quadruped robot include trunk with
And four legs being connected with the trunk, in a cycle of taking a step of quadruped robot, the two legs on same diagonal
Swung in same movement mode, support trunk and promote the two legs referred to as support phase that trunk advances, and according to rail is previously set
Two legs referred to as swing phase on another diagonal that mark is swung, including step is as follows:
A, by analyzing quadruped robot leg structure, build quadruped robot leg exercise Mathematical Modeling, including four-footed
Robot foot end position model:
Every leg includes thigh, shank;One of quadruped robot includes support phase, a swing in cycle of taking a step
Phase, in trunk coordinate system, x-axis represents the displacement in the horizontal direction of sufficient end, and y-axis represents sufficient end highly, and origin o is leg hip joint
Subpoint on the ground, under trunk coordinate system shown in sufficient end position model such as formula (VIII):
In formula (VIII), px(θ1,θ2) it is sufficient end horizontal displacement and θ1、θ2Function, pz(θ1,θ2) it is sufficient end height and θ1、θ2
Function, L1It is the length of robot thigh, L2It is the length of robot shank, θ1It is robot hip joint longitudinal direction folding angle,
Span is 0 ° -180 °;θ2It is knee joint folding angle, span is 0 ° -180 °;H is robot trunk bottom to ground
The distance in face;
B, using a kind of New Sinusoidal diagonal gait, ask for movement locus, pendulum of the support phase foot end under trunk coordinate system
Mutually movement locus of the sufficient end under trunk coordinate system is moved, the New Sinusoidal diagonal gait refers to:Four legs of quadruped robot
Be divided into two groups with diagonal, front left-leg and right rear leg are one group, front right-leg and rear left-leg are one group, two groups alternately as support phase with
Swing phase;Support phase promotes trunk to move ahead, and swing phase is taken a step forward leaping over obstacles, and these processes can be carried out by sufficient end track
Explain.Especially swing phase foot end track, directly determines that swing phase is taken a step in which way.Therefore, sufficient end track
Selection has highly important meaning to motion planning and robot control.
Shown in movement locus such as formula (Ⅸ) of the support phase foot end under trunk coordinate system:
In formula (Ⅸ), p1xT () is support phase foot function of the end horizontal displacement on time t, p1zT () support phase foot end is high
Function of the degree on time t, S is current cycle step length of taking a step, and S_1 is upper one to take a step cycle actual step size, and t is to take a step in the cycle
Moment, T is to take a step the cycle;
Support phase supports trunk and promotes it to move ahead, therefore, to make robot even running, support phase foot end is sat in trunk
The lower motion of mark system should try one's best and remain a constant speed, and keep in the z-axis direction constant.In x0y planes, support phase foot end is carried out at the uniform velocity
Distance do exercises for d=0.5* (S+S_1).
Shown in movement locus such as formula (Ⅹ) of the swing phase foot end under trunk coordinate system:
In formula (Ⅹ), p2xT () is swing phase foot function of the end horizontal displacement on time t, p2zT () swing phase foot end is high
Function of the degree on time t, h is high for step, is the peak reached in swing phase uphill process;
The solution of C, control rate
The span for setting step-length is 0.38m~0.42m, and precision is 1mm, then swing phase foot end is under trunk coordinate system
Movement locus have 41*41=1681 kind possibilities;
Bring 1681 kinds of possibilities corresponding S_1, S into formula (Ⅹ), try to achieve swing phase foot end position p2 not in the same timex
(t)、p2z(t);
Swing phase foot end position p2 not in the same time will be tried to achievex(t)、p2zT () substitutes into the p in formula (VIII)x(θ1,θ2)、pz
(θ1,θ2), a Nonlinear System of Equations is obtained, the Nonlinear System of Equations is solved by MATLAB, and according to joint constraint conditional filtering
Obtain meeting the hip joint longitudinal direction folding angle and knee joint folding angle of joint constraint condition, moved as control swing phase
Control rate;
Bring 1681 kinds of possibilities corresponding S_1, S into formula (Ⅹ), try to achieve support phase foot end position p1 not in the same timex
(t)、p1z(t);
The support phase foot end that will be tried to achieve position p1 not in the same timex(t)、p1zT () substitutes into the p in formula (VIII)x(θ1,θ2)、
pz(θ1,θ2), a Nonlinear System of Equations is obtained, the Nonlinear System of Equations is solved by MATLAB, and according to joint constraint condition sieve
Choosing obtains meeting the hip joint longitudinal direction folding angle and knee joint folding angle of joint constraint condition, is moved as control support phase
Control rate;
D, the control rate according to the swing phase motion asked for, the control rate of support phase motion, set up database
Database uses two-dimensional address pointer, and the first dimension is S_1, and the second dimension is S, and there are 2 controls of 2*200 each address
Matrix processed, respectively support phase joint control rate and swing phase joint control rate;
E, by quick look-up table, realize the sufficient end TRAJECTORY CONTROL to quadruped robot
A, ask for S and S_1;
B, according to S_1 and S, carry out fast zoom table, the support phase for having support phase joint control rate is obtained from database
Control matrix and the swing phase control matrix for having swing phase joint control rate;
C, will support phase control matrix algebraic eqation to support phase, phase control matrix algebraic eqation will be swung to swing phase, according to corresponding
Control rate motion.
According to currently preferred, the course changing control to quadruped robot is realized, including step is as follows:
F, the trunk at quadruped robot previous end cycle moment of taking a step is obtained towards yθ_ 1, ideally take a step the cycle
At the end of trunk towards θ;
G, the steering angle Δ θ that trunk is tried to achieve by formula (Ⅺ), formula (Ⅺ) are as follows:
yθ=yθ_1+Δθ(Ⅺ)
H, in robot kinematics, the steering of trunk is turned in the opposite direction under trunk coordinate system by support phase
Move identical angle and realize.Therefore, on the premise of known target steering angle, so that it may the phase that is supported hip joint transverse direction
The control rate at folding angle, the control rate θ 1 at the transversely opened and closed angle of support phase hip joint is tried to achieve by formula (Ⅻ)4:
I, the transversely opened and closed angle in hip joint of swing phase will be returned just, during SBR, y_ θ _ 1=y_ θ _ 2=0, to formula
(XIII) it is Δ θ _ 1 to be iterated computing and try to achieve the positive-angle of returning of swing phase, and size is current trunk towards y_ θ _ 1 and the last week
Difference of the phase initial time trunk towards y_ θ _ 2:
Δ θ _ 1=yθ_1-yθ_2(XIII)
J, the control rate θ 2 that the transversely opened and closed angle of swing phase hip joint is tried to achieve by formula (XIV)4:
K, by control rate distribute, by signal transmission give corresponding leg group, make its according to control rate change, realize four-footed machine
The course changing control of device people.
Compared to sufficient end TRAJECTORY CONTROL, course changing control is relatively simple, and control rate can be tried to achieve directly by equation, it is only necessary to will be controlled
Rate processed accurately realizes the rotary transform tensor of course changing control and swing phase by distributing to corresponding leg group.
According to currently preferred, when quadruped robot is disturbed, quadruped robot attitude is adjusted, it is ensured that four-footed machine
Device people is stable, including step is as follows:
When robot is moved in complicated ground, at the end of taking a step, swing phase foot end may occur at random with ground
It is relative to slide.This can influence the stability of robot motion.In T/10 after end cycle of taking a step, sufficient end TRAJECTORY CONTROL unit with
Turning control cell cooperates and adjusts robot pose, it is assumed that the random disturbance size for occurring is (Δ px,Δpy), (Δ px,
Δpy) refer to compared to the departure of target drop point, including:
It is close to the ground as the leg of swing phase and move forward, while attitude is adjusted, energy consumption is preferably minimized, step-length is
Δ r, is asked for by formula (XV):
At the same time, support phase pushes ahead robot trunkThe transversely opened and closed angle of hip joint changes Δ θ simultaneously2, lead to
Formula (XVI) is crossed to ask for:
By above-mentioned adjustment, although the direction and position of centre of gravity of robot trunk may all change.But left and right leg hip
Joint line is maintained within the middle of support phase foot end drop point and swing phase foot end drop point, and the leg attitude in same leg group
It is identical.Motion process can be so set to possess high stability.
According to currently preferred, the trunk of quadruped robot towards y is asked forθ, trunk position of centre of gravity (yx,yy), ask for public affairs
Shown in formula such as formula (XVII):
In formula (XVII), UθRefer to actual trunk steering angle, UrIt refer to actual quadruped robot displacement of center of gravity distance.
Beneficial effects of the present invention are:
1st, the present invention carries out robot motion's TRAJECTORY CONTROL using PD neutral nets.Compared to traditional control method, PSO-
PD neutral nets have more preferable Nonlinear Decoupling control ability, and control is accurate, and stability is strong, and with preferably anti-interference
Ability.
2nd, the present invention substitutes traditional BP learning algorithm using PSO particle swarm optimization algorithms are improved, due to proposed by the present invention
Existence feedback and the action function that can discontinuously lead towards in the PD neutral nets of quadruped robot movement locus control, because
This BP learning algorithm can not realize the adaptive learning of neutral net, therefore using improvement PSO particle swarm optimization algorithms as god
Through the adaptive learning algorithm of network.The algorithm the convergence speed is fast, high precision, and is difficult to be absorbed in local optimum, can be preferable
The adaptive learning of the PD neutral nets is realized, in sum, proposed by the present invention based on improvement PSO-PD neutral nets four
Biped robot control method of motion trace high speed, accurate, stabilization.
Brief description of the drawings
Fig. 1 is the movement locus and target drop point schematic diagram of quadruped robot;
Fig. 2 is the trunk and leg structure schematic diagram of quadruped robot;
Fig. 3 is swing phase foot end track schematic diagram;
Fig. 4 is the schematic process flow diagram of sufficient end TRAJECTORY CONTROL;
Fig. 5 is the schematic process flow diagram of course changing control;
Fig. 6 is the schematic network structure of PD neutral nets;
Fig. 7 is the control effect figure of PD neutral nets;
Fig. 8 is control effect figure when there is random perturbation.
Specific embodiment
The present invention is further qualified with reference to Figure of description and embodiment, but not limited to this.
Embodiment
A kind of quadruped robot control method of motion trace based on PSO-PD neutral nets, including:
(1) according to quadruped robot movement locus and default step-length, quadruped robot center of gravity is asked on the movement locus
Target drop point, i.e. coordinate of the quadruped robot center of gravity on the movement locus;It is for actual result.
(2) target drop point, the body of Real-time Feedback of the quadruped robot center of gravity for asking for step (1) on the movement locus
Dry position of centre of gravity (yx,yy) it is input into the input layer of PD neural network models;Into the first of PD neural network models after linear transformation
Hidden layer, carries out scale operation and differentiates in the first hidden layer;Again into PD neural network models after linear transformation
Second hidden layer, the displacement of instructing on x directions and y directions is obtained in the second hidden layer, and residing coordinate system is that quadruped robot is transported
The coordinate system of the horizontal plane where dynamic, origin is quadruped robot original center of gravity position;Into the of PD neural network models
Three hidden layers, try to achieve in the 3rd hidden layer and instruct shift length dr, instruct steering angle d θ, and it refers to four-footed to instruct shift length dr
The distance that robot is stepped from current this step to next step, instruct steering angle refer to quadruped robot from current this step to
The trunk steering angle of next step;Into the layer 5 of PD neural network models, layer 5 is respectively sufficient end TRAJECTORY CONTROL unit
With turning control cell.Sufficient end TRAJECTORY CONTROL unit is received and instructed distance and the actual displacement distance in a upper cycle, realizes four-footed
The sufficient end TRAJECTORY CONTROL of robot;Turning control cell is received and instructs trunk direction and current trunk direction, realizes four-footed machine
The course changing control of people, sufficient end TRAJECTORY CONTROL, the course changing control to quadruped robot are realized in layer 5;Into PD neutral net moulds
The layer 6 of type, when quadruped robot foot end is slided, adjusts quadruped robot attitude, it is ensured that quadruped robot operation is steady
It is fixed;But the process can make robot center of gravity change, the random disturbance of the size that random times occur is can be considered, in addition with
The disturbance that many other reasonses are caused, adds disturbance to be simulated and will add the step-length after disturbing as actual step in layer 6
Length feeds back to layer 5;Into the layer 7 of PD neural network models, the trunk of quadruped robot towards y is asked forθ, trunk center of gravity
Position (yx,yy), trunk is towards yθIt refer to the angle of the actual direction of trunk, i.e. trunk and X-axis after currently taking a step to terminate;Trunk
Position of centre of gravity (yx,yy) refer to coordinate of the quadruped robot center of gravity in the coordinate system;The trunk of quadruped robot is towards yθInstead
Feed the layer 5 of PD neural network models, trunk position of centre of gravity (yx,yy) feed back to the input layer of PD neural network models.Should
The structure of PD neural network models is as shown in Figure 6;
Using the connection weight between multilayer neuron in improved particle swarm optimization algorithm adjustment PD neural network models,
The adaptive learning of the PD neutral nets of object manipulator movement locus control is realized, including:
A, the On The Choice of the connection weight between multilayer neuron is converted into optimization problem, the mesh of optimization problem
Scalar functions are the L2 norms of output vector and tutor's signal vector, and as shown in formula (I), tutor's signal is target landing point coordinates, defeated
Outgoing vector is quadruped robot actual center gravity position;
In formula (I), Error is the object function of optimization problem, and xd (k), yd (k) are respectively quadruped robot kth step
Center of gravity target drop point abscissa, ordinate, yx(k)、yyK () is the reality of the center of gravity target drop point of quadruped robot kth step
Abscissa, ordinate;
B, the span that each connection weight is determined using previous experience, that is, determine Search Range;
C, random initializtion a group particle, i.e. population in Search Range, including initialization particle initial position with
Initial velocity, with these three index expression particle characteristicses of position, speed and fitness, institute in positional representation PD neural network models
Some connection weight values, speed represents the direction of each particle evolution, and fitness value is tried to achieve by fitness function, i.e. each grain
The corresponding object function of son;And each particle represents a potential optimal solution of above-mentioned optimization problem;
The speed of particle is according to the current location of particle, present speed, the history optimum position Pbest of particle and population
The position Gbest of middle optimal particle updates, the speed of particleMore new formula such as formula (II) shown in:
In formula (II), id is the numbering of particle in population,It is the speed of the i-th generation particle,For the i-th generation particle exists
I-th instead of preceding history optimum position,It is the position of optimal particle in the i-th generation population;ω (i) is the used of the i-th generation particle
Property weight, its size determines that speed to what extent inherits the movement velocity of previous generation particles;%1, %2It is acceleration factor,
Value is nonnegative constant;r1、r2It is the random number between 0 to 1;It is the position of the i-th generation particle;The value of ω during initialization
ωstartIt is 0.9, the value ω of ω at the end of iteration34dIt is 0.01, inertia weight ω accelerates decay in an iterative process, the initial stage is excellent
First speed of searching optimization, the later stage focuses on low optimization accuracy, shown in the more new formula such as formula (III) of inertia weight ω:
In formula (III), maxg;N is maximum iteration;
Present invention adds decreases in non-linear inertia weight ω (i), inertia weight ω (i) determines speed to what extent
The movement velocity of previous generation is inherited, the bigger speed of searching optimization of ω (i) is faster in an iterative process, but precision is lower;Otherwise then precision
Higher, speed is lower.To reach the balance of speed and precision, inertia weight ω (i) is larger at the iteration initial stage, as iteration is carried out
Inertia weight progressively reduces.
The speed of the particle after being updatedAfterwards, the position of the particle, the position of particle are updatedRenewal it is public
Shown in formula such as formula (IV):
If formula (IV) ask forCorresponding object function is less thanCorresponding fitness function, then
Conversely,Meanwhile, the position of optimal particle, obtains in Population Regeneration
In this way, carrying out successive ignition obtains approximate optimal solution, even if PD neural network models control error minimum is near
Like best initial weights.
After adding the thought of the TSP question in genetic algorithm, i.e. particle position to update, there is certain probability to occur
Variation in span.TSP question has expanded in iteration the population search space for constantly reducing, can be
Bigger space development search, makes it jump out local optimum, and increase obtains the possibility of global optimum.Each particle in position more
After new, a random number can be generated, if greater than 0.9 random initializtion particle again in Search Range.If less than
Doed nothing equal to 0.9.
Improved particle swarm optimization algorithm takes into account convergence rate and low optimization accuracy while being difficult to be absorbed in local optimum.With
It substitutes the adaptive learning that traditional PS O particle cluster algorithms realize PSO-PD neutral nets.Improved particle swarm optimization algorithm, learns
The habit time is shorter, and control accuracy is higher, therefore can better meet the demand for control of quadruped robot movement locus control.
Step (1), according to quadruped robot movement locus, asks for target of the quadruped robot on the movement locus and falls
Point, the movement locus of quadruped robot is sine curve in the present embodiment, during along sinusoidal motion, the turning half of robot
Footpath, angle of turn can be used to prove whether the control strategy is widely used in most non-all constantly changing
Linear motion trajectory.It is 0.4m to set default step-length L, and when not having external disturbance, the center of gravity of robot will not change, machine
The forward travel distance of device people is step-length;Ask for formula such as formula (V) institute of target drop point of the quadruped robot on the movement locus
Show:
In formula (V), x_1, y_1 are respectively the abscissa and ordinate of back quadruped robot center of gravity, and x, y are respectively
When the abscissa and ordinate of back quadruped robot center of gravity;
The initial value of x_1, y_1 is 0, after asking for x, y, then is assigned to x_1, y_1, iterates after 200 times, constitutes
One matrix of 200*2, obtains target drop point of the quadruped robot on the movement locus.As shown in Figure 1.
Tried to achieve in the 3rd hidden layer and instruct shift length dr, instruct trunk towards d θ, including:
Ask for instructing shift length dr by formula (VI), formula (VI) is as follows:
In formula (VI), d Δs x, d Δ y is instructed x increments and is instructed y increments, i.e. x directions and y by what the second hidden layer was tried to achieve
Displacement is instructed on direction;
Ask for instructing trunk towards d θ by formula (VII), formula (VII) is as follows:
Realize to the sufficient end TRAJECTORY CONTROL of quadruped robot, quadruped robot includes trunk and is connected with the trunk
Four legs, in a cycle of taking a step of quadruped robot, the two legs on same diagonal are swung in same movement mode,
Support trunk simultaneously promotes the two legs referred to as support phase that trunk advances, and another diagonal according to be previously set that track swung
Two legs on line are referred to as swing phase, including step is as follows:
A, by analyzing quadruped robot leg structure, build quadruped robot leg exercise Mathematical Modeling, including four-footed
Robot foot end position model:
Every leg includes thigh, shank;One of quadruped robot includes support phase, a swing in cycle of taking a step
Phase, in trunk coordinate system, x-axis represents the displacement in the horizontal direction of sufficient end, and y-axis represents sufficient end highly, and origin o is leg hip joint
Subpoint on the ground, under trunk coordinate system shown in sufficient end position model such as formula (VIII):
In formula (VIII), px(θ1,θ2) it is sufficient end horizontal displacement and θ1、θ2Function, pz(θ1,θ2) it is sufficient end height and θ1、θ2
Function, the trunk of quadruped robot is with leg structure schematic diagram as shown in Fig. 2 L1It is the length of robot thigh, L1=
300mm, L2It is the length of robot shank, L2=300mm, θ1It is robot hip joint longitudinal direction folding angle, span is
0°-180°;θ2It is knee joint folding angle, span is 0 ° -180 °;H is distance of the robot trunk bottom to ground;
B, using a kind of New Sinusoidal diagonal gait, ask for movement locus, pendulum of the support phase foot end under trunk coordinate system
Mutually movement locus of the sufficient end under trunk coordinate system is moved, the New Sinusoidal diagonal gait refers to:Four legs of quadruped robot
Be divided into two groups with diagonal, front left-leg and right rear leg are one group, front right-leg and rear left-leg are one group, two groups alternately as support phase with
Swing phase;Support phase promotes trunk to move ahead, and swing phase is taken a step forward leaping over obstacles, and these processes can be carried out by sufficient end track
Explain.Especially swing phase foot end track, directly determines that swing phase is taken a step in which way.Therefore, sufficient end track
Selection has highly important meaning to motion planning and robot control.
Shown in movement locus such as formula (Ⅸ) of the support phase foot end under trunk coordinate system:
In formula (Ⅸ), p1xT () is support phase foot function of the end horizontal displacement on time t, p1zT () support phase foot end is high
Function of the degree on time t, S is current cycle step length of taking a step, and S_1 is upper one to take a step cycle actual step size, and t is to take a step in the cycle
Moment, T is to take a step the cycle;
Support phase supports trunk and promotes it to move ahead, therefore, to make robot even running, support phase foot end is sat in trunk
The lower motion of mark system should try one's best and remain a constant speed, and keep in the z-axis direction constant.In x0y planes, support phase foot end is carried out at the uniform velocity
Distance do exercises for d=0.5* (S+S_1).
Shown in movement locus such as formula (Ⅹ) of the swing phase foot end under trunk coordinate system:
In formula (Ⅹ), p2xT () is swing phase foot function of the end horizontal displacement on time t, p2zT () swing phase foot end is high
Function of the degree on time t, h is high for step, is the peak reached in swing phase uphill process;Swing phase foot end track schematic diagram
As shown in Figure 3.
The solution of C, control rate
The span for setting step-length is 0.38m~0.42m, and precision is 1mm, then swing phase foot end is under trunk coordinate system
Movement locus have 41*41=1681 kind possibilities;
Bring 1681 kinds of possibilities corresponding S_1, S into formula (Ⅹ), try to achieve swing phase foot end position p2 not in the same timex
(t)、p2z(t);
Swing phase foot end position p2 not in the same time will be tried to achievex(V)、p2zT () substitutes into the p in formula (VIII)x(θ1,θ2)、pz
(θ1,θ2), a Nonlinear System of Equations is obtained, the Nonlinear System of Equations is solved by MATLAB, and according to joint constraint conditional filtering
Obtain meeting the hip joint longitudinal direction folding angle and knee joint folding angle of joint constraint condition, moved as control swing phase
Control rate;
Bring 1681 kinds of possibilities corresponding S_1, S into formula (Ⅹ), try to achieve support phase foot end position p1 not in the same timex
(t)、p1z(t);
The support phase foot end that will be tried to achieve position p1 not in the same timex(t)、p1zT () substitutes into the p in formula (VIII)x(θ1,θ2)、
pz(θ1,θ2), a Nonlinear System of Equations is obtained, the Nonlinear System of Equations is solved by MATLAB, and according to joint constraint condition sieve
Choosing obtains meeting the hip joint longitudinal direction folding angle and knee joint folding angle of joint constraint condition, is moved as control support phase
Control rate;
D, the control rate according to the swing phase motion asked for, the control rate of support phase motion, set up database
Database uses two-dimensional address pointer, and the first dimension is S_1, and the second dimension is S, and there are 2 controls of 2*200 each address
Matrix processed, respectively support phase joint control rate and swing phase joint control rate;
E, by quick look-up table, realize the sufficient end TRAJECTORY CONTROL to quadruped robot, as shown in Figure 4:
A, ask for S and S_1;
B, according to S_1 and S, carry out fast zoom table, the support phase for having support phase joint control rate is obtained from database
Control matrix and the swing phase control matrix for having swing phase joint control rate;
C, will support phase control matrix algebraic eqation to support phase, phase control matrix algebraic eqation will be swung to swing phase, according to corresponding
Control rate motion.
The course changing control to quadruped robot is realized, as shown in figure 5, as follows including step:
F, the trunk at quadruped robot previous end cycle moment of taking a step is obtained towards yθ_ 1, ideally take a step the cycle
At the end of trunk towards θ;
G, the steering angle Δ θ that trunk is tried to achieve by formula (Ⅺ), formula (Ⅺ) are as follows:
yθ=yθ_1+Δθ(Ⅺ)
H, in robot kinematics, the steering of trunk is turned in the opposite direction under trunk coordinate system by support phase
Move identical angle and realize.Therefore, on the premise of known target steering angle, so that it may the phase that is supported hip joint transverse direction
The control rate at folding angle, the control rate θ 1 at the transversely opened and closed angle of support phase hip joint is tried to achieve by formula (Ⅻ)4:
I, the transversely opened and closed angle in hip joint of swing phase will be returned just, during SBR, y_ θ _ 1=y_ θ _ 2=0, to formula
(XIII) it is Δ θ _ 1 to be iterated computing and try to achieve the positive-angle of returning of swing phase, and size is current trunk towards y_ θ _ 1 and the last week
Difference of the phase initial time trunk towards y_ θ _ 2:
Δ θ _ 1=yθ_1-yθ_2(XIII)
J, the control rate θ 2 that the transversely opened and closed angle of swing phase hip joint is tried to achieve by formula (XIV)4:
K, by control rate distribute, by signal transmission give corresponding leg group, make its according to control rate change, realize four-footed machine
The course changing control of device people.
Compared to sufficient end TRAJECTORY CONTROL, course changing control is relatively simple, and control rate can be tried to achieve directly by equation, it is only necessary to will be controlled
Rate processed accurately realizes the rotary transform tensor of course changing control and swing phase by distributing to corresponding leg group.
When quadruped robot is disturbed, quadruped robot attitude is adjusted, it is ensured that quadruped robot is stable, including
Step is as follows:
When robot is moved in complicated ground, at the end of taking a step, swing phase foot end may occur at random with ground
It is relative to slide.This can influence the stability of robot motion.In T/10 after end cycle of taking a step, sufficient end TRAJECTORY CONTROL unit with
Turning control cell cooperates and adjusts robot pose, it is assumed that the random disturbance size for occurring is (Δ px,Δpy), (Δ px,
Δpy) refer to compared to the departure of target drop point, including:
It is close to the ground as the leg of swing phase and move forward, while attitude is adjusted, energy consumption is preferably minimized, step-length is
Δ r, is asked for by formula (XV):
At the same time, support phase pushes ahead robot trunkThe transversely opened and closed angle of hip joint changes Δ θ simultaneously2,
Asked for by formula (XVI):
By above-mentioned adjustment, although the direction and position of centre of gravity of robot trunk may all change.But left and right leg hip
Joint line is maintained within the middle of support phase foot end drop point and swing phase foot end drop point, and the leg attitude in same leg group
It is identical.Motion process can be so set to possess high stability.
Ask for the trunk of quadruped robot towards yθ, trunk position of centre of gravity (yx,yy), ask for shown in formula such as formula (XVII):
In formula (XVII), UθRefer to actual trunk steering angle, UrIt refer to actual quadruped robot displacement of center of gravity distance.
PD neutral nets (are added without disturbing in the case where sufficient end does not occur deviation and no other disturbances in simulation process
In the case of dynamic) control effect figure as shown in fig. 7, the accumulated error of 200 steps is 0.1360m, actual center gravity after operation program
Drop point is essentially coincided with target barycentric drop point, control effect fast accurate.
Swing phase foot end the random slip of size can occur after unstructuredness ground is touched, due to step-length restriction most
Big step-length is 0.42m, and minimum step is 0.38m, therefore, the maximum sliding distance in x-axis, y-axis is 0.04m, to machine
Maximum perturbation of people's center of gravity in x-axis, y-axis is 0.02m, it is contemplated that the disturbance that other reasonses cause, by robot center of gravity in x
Maximum perturbation in axle, y-axis is set to 0.08m, and the random generation of disturbance, and the probability of generation is 10%, with the anti-of test system
Interference performance.By program verification, the accumulation disturbance of 200 steps is 1.5544m, and the accumulated error of control system is 1.8694m,
As shown in figure 8, after generation is disturbed, control system can immediately be reduced or eliminated error.It can be seen that based on PSO-PD neutral nets
Quadruped robot control is accurate, strong antijamming capability, with high practical value.
Claims (9)
1. a kind of quadruped robot control method of motion trace based on PSO-PD neutral nets, it is characterised in that including:
(1) according to quadruped robot movement locus and default step-length, mesh of the quadruped robot center of gravity on the movement locus is asked for
Village point, i.e. coordinate of the quadruped robot center of gravity on the movement locus;
(2) target drop point, the trunk weight of Real-time Feedback of the quadruped robot center of gravity for asking for step (1) on the movement locus
Heart position (yx,yy) it is input into the input layer of PD neural network models;Implied into the first of PD neural network models after linear transformation
Layer, carries out scale operation and differentiates in the first hidden layer;Again into the second of PD neural network models after linear transformation
Hidden layer, the displacement of instructing on x directions and y directions is obtained in the second hidden layer, and residing coordinate system is that quadruped robot moves institute
Horizontal plane coordinate system, origin be quadruped robot original center of gravity position;Into PD neural network models the 3rd is hidden
Containing layer, tried to achieve in the 3rd hidden layer and instruct shift length dr, instruct steering angle d θ, it refers to four-footed machine to instruct shift length dr
The distance that people steps from current this step to next step, it refers to quadruped robot from current this step to next to instruct steering angle
The trunk steering angle of step;Into the layer 5 of PD neural network models, the sufficient end rail to quadruped robot is realized in layer 5
Mark control, course changing control;Into the layer 6 of PD neural network models, when quadruped robot foot end is slided, adjustment four
Biped robot attitude, it is ensured that quadruped robot is stable;Layer 6 add disturbance be simulated, and will add disturbance after
Step-length feeds back to layer 5 as actual step size;Into the layer 7 of PD neural network models, the trunk of quadruped robot is asked for
Towards yθ, trunk position of centre of gravity (yx,yy), trunk is towards yθRefer to the actual direction of trunk, i.e. trunk after currently taking a step to terminate
With the angle of X-axis;Trunk position of centre of gravity (yx,yy) refer to coordinate of the quadruped robot center of gravity in the coordinate system;Four-footed machine
The trunk of people is towards yθFeed back to the layer 5 of PD neural network models, trunk position of centre of gravity (yx,yy) feed back to PD neutral nets
The input layer of model.
2. a kind of quadruped robot control method of motion trace based on PSO-PD neutral nets according to claim 1,
Characterized in that, using the connection weight between multilayer neuron in improved particle swarm optimization algorithm adjustment PD neural network models
Value, realizes the adaptive learning of the PD neutral nets of object manipulator movement locus control, including:
A, the On The Choice of the connection weight between multilayer neuron is converted into optimization problem, the target letter of optimization problem
Number is the L2 norms of output vector and tutor's signal vector, and as shown in formula (I), tutor's signal is target landing point coordinates, export to
Amount is quadruped robot actual center gravity position;
In formula (I), Error is the object function of optimization problem, and xd (k), yd (k) are respectively the weight of quadruped robot kth step
The abscissa of target centroid drop point, ordinate, yx(k)、yyK () is the actual horizontal seat of the center of gravity target drop point of quadruped robot kth step
Mark, ordinate;
B, the span that each connection weight is determined using previous experience, that is, determine Search Range;
C, random initializtion a group particle, i.e. population in Search Range, including initialization particle initial position with it is initial
Speed, it is all of in positional representation PD neural network models with these three index expression particle characteristicses of position, speed and fitness
Connection weight value, speed represents the direction of each particle evolution, and fitness value is tried to achieve by fitness function, i.e. each particle pair
The object function answered;
The speed of particle according in the current location of particle, present speed, the history optimum position Pbest of particle and population most
The position Gbest of excellent particle updates, the speed of particleMore new formula such as formula (II) shown in:
In formula (II), id is the numbering of particle in population,It is the speed of the i-th generation particle,It is the i-th generation particle in the i-th generation
History optimum position before,It is the position of optimal particle in the i-th generation population;ω (i) is the inertia power of the i-th generation particle
Weight, its size determines that speed to what extent inherits the movement velocity of previous generation particles;%1, %2It is acceleration factor, value
It is nonnegative constant;r1、r2It is the random number between 0 to 1;It is the position of the i-th generation particle;The value ω of ω during initializationstart
It is 0.9, the value ω of ω at the end of iteration34dIt is 0.01, inertia weight ω accelerates decay in an iterative process, the initial stage preferentially seeks
Excellent speed, the later stage focuses on low optimization accuracy, shown in the more new formula such as formula (III) of inertia weight ω:
In formula (III), maxgen is maximum iteration;
The speed of the particle after being updatedAfterwards, the position of the particle, the position of particle are updatedMore new formula such as formula
(IV) shown in:
If formula (IV) ask forCorresponding object function is less thanCorresponding fitness function, thenConversely,Meanwhile, the position of optimal particle, obtains in Population Regeneration
In this way, carrying out successive ignition obtains approximate optimal solution, though PD neural network models control error it is minimum it is approximate most
Excellent weights.
3. a kind of quadruped robot control method of motion trace based on PSO-PD neutral nets according to claim 1,
Characterized in that, the step (1), according to quadruped robot movement locus, quadruped robot is asked on the movement locus
Target drop point, the movement locus for setting quadruped robot is sine curve, and the span for setting default step-length L is 0.38-
0.42m, asks for shown in the formula such as formula (V) of target drop point of the quadruped robot on the movement locus:
In formula (V), x_1, y_1 are respectively the abscissa and ordinate of back quadruped robot center of gravity, and x, y are respectively currently
The abscissa and ordinate of one step quadruped robot center of gravity;
The initial value of x_1, y_1 is 0, after asking for x, y, then is assigned to x_1, y_1, after iterating, obtains quadruped robot
Target drop point on the movement locus.
4. a kind of quadruped robot control method of motion trace based on PSO-PD neutral nets according to claim 3,
Characterized in that, the value of L is 0.4m.
5. a kind of quadruped robot control method of motion trace based on PSO-PD neutral nets according to claim 1,
Characterized in that, tried to achieve in the 3rd hidden layer and instruct shift length dr, instruct trunk towards d θ, including:
Ask for instructing shift length dr by formula (VI), formula (VI) is as follows:
In formula (VI), d Δs x, d Δ y is instructed x increments and is instructed y increments, i.e. x directions and y directions by what the second hidden layer was tried to achieve
On instruct displacement;
Ask for instructing trunk towards d θ by formula (VII), formula (VII) is as follows:
6. a kind of quadruped robot control method of motion trace based on PSO-PD neutral nets according to claim 1,
Characterized in that, realizing the sufficient end TRAJECTORY CONTROL to quadruped robot, quadruped robot includes trunk and connects with the trunk
Four legs for connecing, in a cycle of taking a step of quadruped robot, the two legs on same diagonal are in same movement mode
Swing, support trunk and promote the two legs referred to as support phase that trunk advances, and it is another according to be previously set that track swung
Two legs on a pair of linea angulatas are referred to as swing phase, including step is as follows:
A, by analyzing quadruped robot leg structure, build quadruped robot leg exercise Mathematical Modeling, including four-footed machine
People's foot end position model:
Every leg includes thigh, shank;One of quadruped robot takes a step the cycle including support phase, a swing phase, body
In dry coordinate system, x-axis represents the displacement in the horizontal direction of sufficient end, and y-axis represents sufficient end highly, and origin o is leg hip joint on ground
Subpoint on face, under trunk coordinate system shown in sufficient end position model such as formula (VIII):
In formula (VIII), px(θ1,θ2) it is sufficient end horizontal displacement and θ1、θ2Function, pz(θ1,θ2) it is sufficient end height and θ1、θ2Letter
Number, L1It is the length of robot thigh, L2It is the length of robot shank, θ1It is robot hip joint longitudinal direction folding angle, value
Scope is 0 ° -180 °;θ2It is knee joint folding angle, span is 0 ° -180 °;H is robot trunk bottom to ground
Distance;
B, using a kind of New Sinusoidal diagonal gait, ask for movement locus, swing phase of the support phase foot end under trunk coordinate system
Movement locus of the sufficient end under trunk coordinate system, the New Sinusoidal diagonal gait refers to:Four legs of quadruped robot are with right
Angle is divided into two groups, and front left-leg and right rear leg are one group, and front right-leg and rear left-leg are one group, and two groups alternately as support phase and swing
Phase;Support phase promotes trunk to move ahead, and swing phase is taken a step forward leaping over obstacles;
Shown in movement locus such as formula (Ⅸ) of the support phase foot end under trunk coordinate system:
In formula (Ⅸ), p1xT () is support phase foot function of the end horizontal displacement on time t, p1zHighly close at (t) support phase foot end
In the function of time t, S is current cycle step length of taking a step, and S_1 takes a step cycle actual step size for upper one, t be in the cycle of taking a step when
Carve, T is to take a step the cycle;
Shown in movement locus such as formula (Ⅹ) of the swing phase foot end under trunk coordinate system:
In formula (Ⅹ), p2xT () is swing phase foot function of the end horizontal displacement on time t, p2zHighly close at (t) swing phase foot end
In the function of time t, h is high for step, is the peak reached in swing phase uphill process;
The solution of C, control rate
The span for setting step-length is 0.38m~0.42m, and precision is 1mm, then fortune of the swing phase foot end under trunk coordinate system
Dynamic rail mark has 41*41=1681 kind possibilities;
Bring 1681 kinds of possibilities corresponding S_1, S into formula (Ⅹ), try to achieve swing phase foot end position p2 not in the same timex(t)、
p2z(t);
Swing phase foot end position p2 not in the same time will be tried to achievex(t)、p2zT () substitutes into the p in formula (VIII)x(θ1,θ2)、pz(θ1,
θ2), a Nonlinear System of Equations is obtained, the Nonlinear System of Equations is solved by MATLAB, and obtain according to joint constraint conditional filtering
To the hip joint longitudinal direction folding angle and knee joint folding angle that meet joint constraint condition, as the control that control swing phase is moved
Rate processed;
Bring 1681 kinds of possibilities corresponding S_1, S into formula (Ⅹ), try to achieve support phase foot end position p1 not in the same timex(t)、
p1z(t);
The support phase foot end that will be tried to achieve position p1 not in the same timex(t)、p1zT () substitutes into the p in formula (VIII)x(θ1,θ2)、pz
(θ1,θ2), a Nonlinear System of Equations is obtained, the Nonlinear System of Equations is solved by MATLAB, and according to joint constraint conditional filtering
Obtain meeting the hip joint longitudinal direction folding angle and knee joint folding angle of joint constraint condition, moved as control support phase
Control rate;
D, the control rate according to the swing phase motion asked for, the control rate of support phase motion, set up database
Database uses two-dimensional address pointer, and the first dimension is S_1, and the second dimension is S, and there are 2 control squares of 2*200 each address
Battle array, respectively support phase joint control rate and swing phase joint control rate;
E, by quick look-up table, realize the sufficient end TRAJECTORY CONTROL to quadruped robot
A, ask for S and S_1;
B, according to S_1 and S, carry out fast zoom table, the support phase control for having support phase joint control rate is obtained from database
Matrix and the swing phase control matrix for having swing phase joint control rate;
C, will support phase control matrix algebraic eqation to support phase, phase control matrix algebraic eqation will be swung to swing phase, according to corresponding control
Rate motion processed.
7. a kind of quadruped robot control method of motion trace based on PSO-PD neutral nets according to claim 1,
Characterized in that, the course changing control to quadruped robot is realized, including step is as follows:
F, the trunk at quadruped robot previous end cycle moment of taking a step is obtained towards yθ_ 1, ideally take a step end cycle
When trunk towards θ;
G, the steering angle Δ θ that trunk is tried to achieve by formula (Ⅺ), formula (Ⅺ) are as follows:
yθ=yθ_1+Δθ (Ⅺ)
H, the control rate θ 1 that the transversely opened and closed angle of support phase hip joint is tried to achieve by formula (Ⅻ)4:
I, the transversely opened and closed angle in hip joint of swing phase will be returned just, and during SBR, y_ θ _ 1=y_ θ _ 2=0 enter to formula (XIII)
The positive-angle of returning that swing phase is tried to achieve in row iteration computing is Δ θ _ 1, when size is that current trunk begins towards y_ θ _ 1 and the last week are initial
Carve difference of the trunk towards y_ θ _ 2:
Δ θ _ 1=yθ_1-yθ_2(XIII)
J, the control rate θ 2 that the transversely opened and closed angle of swing phase hip joint is tried to achieve by formula (XIV)4:
K, by control rate distribute, by signal transmission give corresponding leg group, make its according to control rate change, realize quadruped robot
Course changing control.
8. a kind of quadruped robot control method of motion trace based on PSO-PD neutral nets according to claim 1,
Characterized in that, when quadruped robot is disturbed, adjust quadruped robot attitude, it is ensured that quadruped robot is stable,
It is as follows including step:
In T/10 after end cycle of taking a step, sufficient end TRAJECTORY CONTROL unit cooperates with turning control cell and adjusts robot appearance
State, it is assumed that the random disturbance size for occurring is (Δ px,Δpy), (Δ px,Δpy) refer to compared to the departure of target drop point,
Including:
It is close to the ground as the leg of swing phase and move forward, while attitude is adjusted, energy consumption is preferably minimized, step-length is Δ r,
Asked for by formula (XV):
At the same time, support phase pushes ahead robot trunkThe transversely opened and closed angle of hip joint changes Δ θ simultaneously2, by formula
(XVI) ask for:
9. a kind of quadruped robot control method of motion trace based on PSO-PD neutral nets according to claim 6,
Characterized in that, asking for the trunk of quadruped robot towards yθ, trunk position of centre of gravity (yx,yy), ask for formula such as formula (XVII) institute
Show:
In formula (XVII), UθRefer to actual trunk steering angle, UrIt refer to actual quadruped robot displacement of center of gravity distance.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710295939.9A CN106886155B (en) | 2017-04-28 | 2017-04-28 | Four-legged robot motion trajectory control method based on PSO-PD neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710295939.9A CN106886155B (en) | 2017-04-28 | 2017-04-28 | Four-legged robot motion trajectory control method based on PSO-PD neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106886155A true CN106886155A (en) | 2017-06-23 |
CN106886155B CN106886155B (en) | 2020-01-14 |
Family
ID=59183086
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710295939.9A Active CN106886155B (en) | 2017-04-28 | 2017-04-28 | Four-legged robot motion trajectory control method based on PSO-PD neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106886155B (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107871157A (en) * | 2017-11-08 | 2018-04-03 | 广东工业大学 | Data predication method, system and relevant apparatus based on BP and PSO |
CN110262244A (en) * | 2019-07-02 | 2019-09-20 | 武汉科技大学 | A kind of self adaptation straightening method for improving FSRBFD |
CN110275441A (en) * | 2019-07-02 | 2019-09-24 | 武汉科技大学 | A kind of quick self-adapted decoupling control method of PSORBFD |
CN110426959A (en) * | 2019-08-09 | 2019-11-08 | 太原科技大学 | A kind of caterpillar robot control system |
CN110543100A (en) * | 2019-09-27 | 2019-12-06 | 广东海洋大学 | large truss track optimization method based on minimum energy consumption principle |
CN110597267A (en) * | 2019-09-27 | 2019-12-20 | 长安大学 | Local optimal foot drop point selection method for foot type robot |
CN110799308A (en) * | 2017-09-22 | 2020-02-14 | 谷歌有限责任公司 | Determining a control strategy for a robot using noise tolerant structured search |
CN110861084A (en) * | 2019-11-18 | 2020-03-06 | 东南大学 | Four-legged robot falling self-resetting control method based on deep reinforcement learning |
CN111045581A (en) * | 2019-11-22 | 2020-04-21 | 安徽听见科技有限公司 | Page sliding control method, device, equipment and storage medium |
CN112327869A (en) * | 2020-11-19 | 2021-02-05 | 山东大学 | Diagonal gait motion control method of quadruped robot based on joint velocity planning |
CN112904884A (en) * | 2021-01-28 | 2021-06-04 | 歌尔股份有限公司 | Method and device for tracking trajectory of foot type robot and readable storage medium |
CN113427482A (en) * | 2021-03-18 | 2021-09-24 | 长安大学 | Method, system, equipment and storage medium for robot to pass obstacle |
CN114489092A (en) * | 2020-10-26 | 2022-05-13 | 腾讯科技(深圳)有限公司 | Method, device, equipment and medium for controlling motion of foot type robot |
EP4261625A4 (en) * | 2021-01-28 | 2024-05-29 | Tencent Technology (Shenzhen) Company Limited | Robot motion control method and apparatus, and robot and storage medium |
CN118143955A (en) * | 2024-05-11 | 2024-06-07 | 南京师范大学 | Foot track planning method for foot robot based on pneumatic artificial muscle driving |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2009155030A3 (en) * | 2008-05-28 | 2011-03-03 | Georgia Tech Research Corporation | Systems and methods for providing wireless power to a portable unit |
CN102607552A (en) * | 2012-01-11 | 2012-07-25 | 南京航空航天大学 | Industrial robot space grid precision compensation method based on neural network |
CN103487047A (en) * | 2013-08-06 | 2014-01-01 | 重庆邮电大学 | Improved particle filter-based mobile robot positioning method |
CN105527960A (en) * | 2015-12-18 | 2016-04-27 | 燕山大学 | Mobile robot formation control method based on leader-follow |
CN106334283A (en) * | 2016-10-10 | 2017-01-18 | 南京工程学院 | Fire-fighting and rescue robot system and control method |
-
2017
- 2017-04-28 CN CN201710295939.9A patent/CN106886155B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2009155030A3 (en) * | 2008-05-28 | 2011-03-03 | Georgia Tech Research Corporation | Systems and methods for providing wireless power to a portable unit |
CN102607552A (en) * | 2012-01-11 | 2012-07-25 | 南京航空航天大学 | Industrial robot space grid precision compensation method based on neural network |
CN103487047A (en) * | 2013-08-06 | 2014-01-01 | 重庆邮电大学 | Improved particle filter-based mobile robot positioning method |
CN105527960A (en) * | 2015-12-18 | 2016-04-27 | 燕山大学 | Mobile robot formation control method based on leader-follow |
CN106334283A (en) * | 2016-10-10 | 2017-01-18 | 南京工程学院 | Fire-fighting and rescue robot system and control method |
Non-Patent Citations (1)
Title |
---|
朱龙英 等: "《基于粒子群优化神经网络自适应控制算法的并联》", 《组合机床与自动化加工技术》 * |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11697205B2 (en) | 2017-09-22 | 2023-07-11 | Google Llc | Determining control policies for robots with noise-tolerant structured exploration |
CN110799308B (en) * | 2017-09-22 | 2022-09-27 | 谷歌有限责任公司 | Determining a control strategy for a robot using noise tolerant structured search |
CN110799308A (en) * | 2017-09-22 | 2020-02-14 | 谷歌有限责任公司 | Determining a control strategy for a robot using noise tolerant structured search |
CN107871157B (en) * | 2017-11-08 | 2020-06-09 | 广东工业大学 | Data prediction method, system and related device based on BP and PSO |
CN107871157A (en) * | 2017-11-08 | 2018-04-03 | 广东工业大学 | Data predication method, system and relevant apparatus based on BP and PSO |
CN110262244A (en) * | 2019-07-02 | 2019-09-20 | 武汉科技大学 | A kind of self adaptation straightening method for improving FSRBFD |
CN110275441A (en) * | 2019-07-02 | 2019-09-24 | 武汉科技大学 | A kind of quick self-adapted decoupling control method of PSORBFD |
CN110275441B (en) * | 2019-07-02 | 2022-04-12 | 武汉科技大学 | PSORBFD (particle swarm optimization-based adaptive feedback) rapid self-adaptive decoupling control method |
CN110262244B (en) * | 2019-07-02 | 2022-04-01 | 武汉科技大学 | Self-adaptive decoupling control method for improving FSRBFD |
CN110426959B (en) * | 2019-08-09 | 2022-08-26 | 太原科技大学 | Crawler robot control system |
CN110426959A (en) * | 2019-08-09 | 2019-11-08 | 太原科技大学 | A kind of caterpillar robot control system |
CN110543100A (en) * | 2019-09-27 | 2019-12-06 | 广东海洋大学 | large truss track optimization method based on minimum energy consumption principle |
CN110597267A (en) * | 2019-09-27 | 2019-12-20 | 长安大学 | Local optimal foot drop point selection method for foot type robot |
CN110861084A (en) * | 2019-11-18 | 2020-03-06 | 东南大学 | Four-legged robot falling self-resetting control method based on deep reinforcement learning |
CN110861084B (en) * | 2019-11-18 | 2022-04-05 | 东南大学 | Four-legged robot falling self-resetting control method based on deep reinforcement learning |
CN111045581A (en) * | 2019-11-22 | 2020-04-21 | 安徽听见科技有限公司 | Page sliding control method, device, equipment and storage medium |
CN111045581B (en) * | 2019-11-22 | 2022-06-07 | 安徽听见科技有限公司 | Page sliding control method, device, equipment and storage medium |
CN114489092A (en) * | 2020-10-26 | 2022-05-13 | 腾讯科技(深圳)有限公司 | Method, device, equipment and medium for controlling motion of foot type robot |
CN114489092B (en) * | 2020-10-26 | 2023-07-18 | 腾讯科技(深圳)有限公司 | Foot robot motion control method, device, equipment and medium |
CN112327869A (en) * | 2020-11-19 | 2021-02-05 | 山东大学 | Diagonal gait motion control method of quadruped robot based on joint velocity planning |
CN112904884A (en) * | 2021-01-28 | 2021-06-04 | 歌尔股份有限公司 | Method and device for tracking trajectory of foot type robot and readable storage medium |
EP4261625A4 (en) * | 2021-01-28 | 2024-05-29 | Tencent Technology (Shenzhen) Company Limited | Robot motion control method and apparatus, and robot and storage medium |
CN113427482A (en) * | 2021-03-18 | 2021-09-24 | 长安大学 | Method, system, equipment and storage medium for robot to pass obstacle |
CN118143955A (en) * | 2024-05-11 | 2024-06-07 | 南京师范大学 | Foot track planning method for foot robot based on pneumatic artificial muscle driving |
CN118143955B (en) * | 2024-05-11 | 2024-08-06 | 南京师范大学 | Foot track planning method for foot robot based on pneumatic artificial muscle driving |
Also Published As
Publication number | Publication date |
---|---|
CN106886155B (en) | 2020-01-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106886155A (en) | A kind of quadruped robot control method of motion trace based on PSO PD neutral nets | |
CN112936290B (en) | Quadruped robot motion planning method based on layered reinforcement learning | |
CN109483530B (en) | Foot type robot motion control method and system based on deep reinforcement learning | |
CN104932264B (en) | The apery robot stabilized control method of Q learning frameworks based on RBF networks | |
CN108572553B (en) | Motion closed-loop control method of quadruped robot | |
CN107598897A (en) | A kind of method of humanoid robot gait's planning based on human body teaching | |
CN108931988B (en) | Gait planning method of quadruped robot based on central pattern generator, central pattern generator and robot | |
CN113031528B (en) | Multi-legged robot non-structural ground motion control method based on depth certainty strategy gradient | |
CN110045598B (en) | Method for tracking and controlling underground advancing path of cantilever type heading machine | |
CN109760761A (en) | A kind of quadruped robot motion control method based on bionic principle and intuition | |
CN105965506A (en) | Humanoid biped robot walking posture control method based on genetic algorithm | |
CN110989597A (en) | Adaptive path tracking method of integrated fuzzy neural network | |
CN114995479A (en) | Parameter control method of quadruped robot virtual model controller based on reinforcement learning | |
CN106094817B (en) | Intensified learning humanoid robot gait's planing method based on big data mode | |
CN112147889A (en) | Combined obstacle crossing track planning method for quadruped robot | |
Zhao et al. | PID control of vehicle active suspension based on particle Swarm optimization | |
Shao et al. | Recent advances on gait control strategies for hydraulic quadruped robot | |
CN116625369A (en) | Ground robot path planning method based on deep reinforcement learning | |
CN116859975A (en) | Wheel leg robot wheel-foot switching control method based on BP neural network | |
Han et al. | A heuristic gait template planning and dynamic motion control for biped robots | |
Zhang et al. | A survey of vehicle dynamics modeling methods for autonomous racing: Theoretical models, physical/virtual platforms, and perspectives | |
Palmer et al. | Intelligent control of high-speed turning in a quadruped | |
Lee et al. | Time-dependent genetic algorithm and its application to quadruped’s locomotion | |
Yu et al. | Biped robot gait planning based on 3D linear inverted pendulum model | |
CN116237943A (en) | Four-foot robot control method combined with terrain constraint |
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 | ||
CP03 | Change of name, title or address |
Address after: 250353 No. 3501, Daxue Road, Jinan City, Shandong Province Patentee after: Qilu University of Technology (Shandong Academy of Sciences) Country or region after: China Address before: 250353 No. 3501, Daxue Road, Jinan City, Shandong Province Patentee before: Qilu University of Technology Country or region before: China |