CN103336526B - Based on the robot path planning method of coevolution population rolling optimization - Google Patents

Abstract

The invention discloses a kind of robot that can make there is static-obstacle thing and realizing independent navigation under the working environment of Obstacle Position the unknown, adopt many subgroups coevolution particle cluster algorithm to carry out real-time rolling optimization and keep away the robot path planning method hindering process.The present invention has following beneficial effect: can realize robot and have static-obstacle thing and independent navigation under the working environment of Obstacle Position the unknown; The present invention introduces colony's barycenter strategy in the coevolution of many subgroups, improves the search capability of population; Relative to the paths planning method of existing employing standard particle group algorithm (PSO) and coevolution particle cluster algorithm (CPSO), present invention employs survival of the fittest strategy, evolution convergence accuracy and runtime has more superiority; The present invention has to cook up in real time in complex barrier substance environment and effectively keeps away barrier path, has that robustness is good, solution efficiency advantages of higher.

Description

Based on the robot path planning method of coevolution population rolling optimization

Technical field

The invention belongs to the real-time planning technology field of robot path, be specifically related to a kind of robot path planning method based on coevolution population rolling optimization.

Background technology

Path planning refers to by performance requirements such as distance, consuming time or securities, searching machine people from starting point to impact point between a safe clear path, it is the important topic of robot field.According to the understanding to environmental knowledge, path planning can be divided into the path planning two kinds under the path planning of known environment and circumstances not known, the former is also referred to as global path planning; Latter requires that robot utilizes the information in the sensor detection surrounding limited range of self, constantly carries out dynamic path planning, is also referred to as online path planning or local paths planning.Because needs realize keeping away barrier process online, therefore online path planning is much more complicated than global path planning.

Current, local paths planning method mainly contains: Artificial Potential Field Method, rolling window law of planning and intelligent algorithm law of planning etc.Artificial Potential Field Method algorithm structure is simple but there is local extremum, and robot can be absorbed in local extremum and fluctuate; Adopt in rolling window planing method and there is the existing picture of deadlock.In recent years, along with going deep into of studying bionic intelligence, many intelligent optimization algorithms are increasingly being used in robot path planning, as genetic algorithm, ant group algorithm and particle cluster algorithm.

Summary of the invention

In order to overcome the defect existed in prior art, there is provided a kind of robot that can make having static-obstacle thing and realizing independent navigation under the working environment of Obstacle Position the unknown, adopt many subgroups coevolution particle cluster algorithm to carry out real-time rolling optimization and keep away the robot path planning method hindering process.The concrete technical scheme of the present invention is as follows:

Based on the robot path planning method of coevolution population rolling optimization, for planning the walking path of a robot, making some stationary obstruction that robot avoids in working environment, comprising the steps:

Step S1, carries out modeling to working environment;

Step S2, the distance between measuring robots current point and impact point, if distance is less than default distance error, then thinks and exports the terminal collection of each planning by arrival impact point, obtain programme path; Otherwise proceed to step S3;

Step S3, obtains current optimal movement orientation angle according to robot current point and impact point;

Step S4, according to the robot view field's angle preset, divides the GOTO field of robot; According to the motion of the current optimal movement orientation angle determination current robot GOTO field towards place;

Step S5, if clear in the view field of current robot, be that limit sets up local pole coordinate system with current point, the local endpoint of this planning is obtained according to the radius of view preset and current optimal movement orientation angle, and by local endpoint stored in terminal collection, using local endpoint as the current point planned next time, proceed to step S2; If have barrier in the view field of current robot, proceed to step S6;

Step S6, adopts the mode of interative computation to carry out keeping away barrier process based on coevolution particle cluster algorithm; Step S6 comprises further:

Step S61, is divided into several segmentations by radius of view, determine several segment movement orientation angles, segment movement orientation angle is less than field-of-view angle simultaneously; Some particles are set, each particle are encoded to the sequence be made up of some segment movement orientation angles, some particles are formed an initial population, and initialization is carried out to initial population, obtain an initial population; If initialization success, is divided into several subgroups, proceeds to step S62 by initial population; If initialization failure, proceeds to step S7;

Step S62, calculates the adaptive value of each particle in each subgroup, and sorts by adaptive value to the particle in each subgroup, then selects by the gene-ratio preset the particle that in each subgroup, adaptive value is less, forms a gene pool;

Step S63, according to the adaptive value of particle each in each subgroup, obtains the optimum solution of each subgroup, and tries to achieve a population center of mass point according to optimum solution;

Step S64, upgrades speed and the position of particle in each subgroup; If be updated successfully, then proceed to step S65, if upgrade unsuccessfully, proceed to step S7;

Step S65, upgrades the globally optimal solution of the adaptive value of particle in each subgroup, the individual optimal solution of particle and subgroup;

Step S66, judges whether to reach default evolution cycle, if do not reach evolution cycle, directly proceeds to step S67; Otherwise upgrade subgroup and gene pool, then Population Regeneration center of mass point, then proceed to step S67;

Step S67, judges whether to reach default maximum iteration time, if reach maximum iteration time, then termination of iterations, obtains the local endpoint of this planning, and local endpoint is added terminal concentrates, using local endpoint as the current point planned next time, proceed to step S2; If do not reach maximum iteration time, then proceed to step S64 and continue iteration;

Step S7, carries out keeping away barrier failure handling, robot is proceeded to next GOTO field, returns step S61 and carries out keeping away barrier; Hinder unsuccessfully if robot all keeps away at all GOTO fields, then cannot realize keeping away barrier, unsuccessfully jump out.

As prioritization scheme, the working environment modeling of step S1 specifically comprises the steps:

Step S11, ignores the elevation information in working environment, sets up two-dimensional coordinate system;

Step S12, if the view field of robot is (r, θ), wherein r is radius of view, and θ is field-of-view angle; Obtain the starting point S of robot and the coordinate of impact point G; Describe barrier with convex polygon, the coordinate of barrier is unknown;

Step S13, the size according to robot is carried out expanded to barrier, and using robot as a particle process;

Step S14, arranging current scrolling planning number of times t is 1, arranges the current point P of robot ^tfor starting point S, and by current point P ^tstored in terminal collection { P}.

As prioritization scheme, step S3 is specially:

Current optimal movement orientation angle refer to that current kinetic is towards the angle with two-dimensional coordinate system x-axis, current kinetic is oriented current point P ^tto the rectilinear direction of impact point G; Current optimal movement orientation angle computing formula as the formula (1):

(1)

Wherein, t is current scrolling planning number of times, if then make g _xfor described impact point G is at the coordinate figure of x-axis, G _yfor described impact point G is at the coordinate figure of y-axis, P _x ^tfor described current point P ^tat the coordinate figure of x-axis, P _y ^tfor described current point P ^tat the coordinate figure of y-axis.

As prioritization scheme, step S4 is specially:

According to field-of-view angle θ, the GOTO field of robot is divided into K=2 π/θ, according to current optimal movement orientation angle determine the GOTO field k of robot current kinetic towards place, current kinetic towards the GOTO field k at place computing formula as the formula (2):

(2)

Wherein, if k=0, then make k=K, wherein ceil () is the function that rounds up, and mod () is remainder function.

As prioritization scheme, step S5 is specially:

Local endpoint P ^t' computing formula as the formula (3):

(3)

Wherein, P ^t' _xfor described local endpoint P ^t' at the coordinate figure of x-axis, P ^t' _yfor described local endpoint P ^t' at the coordinate figure of y-axis, P _x ^tfor described current point P ^tat the coordinate figure of x-axis, P _y ^tfor described current point P ^tat the coordinate figure of y-axis; By described local endpoint P ^t' { after P}, make t=t+1 stored in described terminal collection.

As prioritization scheme, step S61 comprises the steps: further

Step S611, is divided into m section by radius of view r, obtains a segment radius λ, determines m segment movement orientation angle at random wherein, i=1,2 ... m, obtains some segmented paths dist _i, by some segmented paths dist _iconnect to form the path overall length dist that robot avoiding obstacles within the scope of radius of view r need pass through _t;

Step S612, arranging some particles for carrying out the interative computation of particle cluster algorithm, being encoded to by particle by some segment movement orientation angles the sequence of composition, if the dimension of some particles is m, according to the GOTO field k of current kinetic towards place, defines the scope of some particles for [(k-1) θ, k θ], wherein, k=1,2 ... K, K=2 π/θ;

Step S613, determines path overall length dist _t, path overall length dist _tcomputing formula as the formula (4):

${\mathrm{dist}}_t=\mathrm{\λ}+\underset{i-2}{\overset{m}{\mathrm{\Σ}}}{\mathrm{dist}}_i=r/m+\underset{i=2}{\overset{m}{\mathrm{\Σ}}}\sqrt{{\left((i-1)\mathrm{\λ}\right)}^2+{\left(\mathrm{i\λ}\right)}^2+2\mathrm{i\λ}(i-1)\mathrm{\λ}\cos (x_i-x_{i-1})}$ (4)

Wherein, i=2,3 ..., m; λ=r/m, namely radius of view r makes the section length after m decile; x _ifor the i-th dimension of particle after encoding;

Obtain local endpoint P ^t' to the air line distance of impact point G air line distance computing formula as the formula (5):

${\mathrm{dist}}_{P^{t\′}G}=\sqrt{{({P^{t\′}}_x-G_x)}^2+{\left({P^{t\′}}_y{-G}_y\right)}^2}$ (5)

According to path overall length dist _tand air line distance set an adaptive value function, adaptive value function as the formula (6):

$\mathrm{Fit}={\mathrm{\ω}}_1{\mathrm{dist}}_t+{\mathrm{\ω}}_2{\mathrm{dist}}_{P^{t\′}G}$ (6)

ω wherein ₁with ω ₂for weight coefficient;

Step S614, forms an initial population by some particles, carries out initialization to initial population; The current point P that initialization requirements is planned from this ^tstart, with segmented paths dist _ifor there is not barrier in the view field of radius of view, otherwise initialization failure;

Step S615, if initialization success, is divided into s subgroup, according to the GOTO field k of current kinetic towards place, defines the scope [x of some particles by initial population _min, x _max] be [(k-1) θ, k θ]; The maximal rate arranging particle is 0.1 times of particle range.

As prioritization scheme, the initialization in step S614 is specially:

Setting original dimensions is m, initial scale is N, before carrying out initialization, arranges the population scale of initial population much larger than initial scale N; The maximum initialization times p of particle is set _maxinitialization times d maximum with dimension _max; If certain one dimension of a certain particle is making d in initial population _maxstill unsuccessful after secondary initialization, then reinitialize current dimension; If this particle is through p _maxsecondary initialization is still failed, then think that current particle initialization is failed;

Abandon the particle of initialization failure, by the particle of successful initialization composition initial population; If the scale of initial population reaches initial scale N, then initialization success; Otherwise, initialization failure.

As prioritization scheme, population center of mass point P in step S63 _cdefinition as the formula (7):

$P_c(R+1)=\frac{1}{s}\underset{1}{\overset{s}{\mathrm{\Σ}}}{p}_{\mathrm{gd}}^i\left(R\right)i=\mathrm{1,2},...,s$ (7)

Wherein, R is evolution cycle, it is the globally optimal solution of i-th subgroup.

As prioritization scheme, step S64 is specially:

Speed and the position of particle in each subgroup is upgraded according to formula (8) and formula (9):

${v_i}^j(\mathrm{iter}+1)=\mathrm{\ω}\×{v_i}^j\left(\mathrm{iter}\right)+c_1{r}_1\×({P_i}^j\left(\mathrm{iter}\right)-{x_i}^j\left(\mathrm{iter}\right))+c_2{r}_2\×({g_i}^j\left(\mathrm{iter}\right)-{x_i}^j\left(\mathrm{iter}\right))+c_3{r}_3({P_c}^R-{x_i}^j\left(\mathrm{iter}\right))$

(8)

$x_i^j(\mathrm{iter}+1)=x_i^j\left(\mathrm{iter}\right)+v_i^j(\mathrm{iter}+1)$ (9)

Wherein iter is Evolution of Population algebraically, and j is subgroup mark, and i is particle label in subgroup, for the speed of i-th particle in a jth subgroup; P _i ^jfor the history optimum solution of i-th particle in a jth subgroup; for the optimum solution of subgroup, current particle place; c ₁and c ₂for Studying factors; c ₃for exchanging the factor; r ₁, r ₂, r ₃for the random number between [0,1]; ω is inertia weight, and inertia weight ω adopts linear decrease strategy, as the formula (10):

ω(iter)=ω _max-iter×(ω _max-ω _min)/iter _max(10)

The speed of particle and location boundary process are comprised: if the d of a certain particle i ties up speed in the j of subgroup $v_{i,j}^d>v_{\max },$ Then make $v_{i,j}^d=v_{\max };$ If $v_{i,j}^d<v_{\min },$ Then make $v_{i,j}^d=v_{\min };$ If $x_{i,j}^d>x_{\max },$ Then make $x_{i,j}^d=x_{\max };$ If $x_{i,j}^d<x_{\min },$ Then $x_{i,j}^d=x_{\min };$

Each dimension after particle position upgrades is required to include: from current point P ^tstart, with segmented paths dist _ifor there is not barrier in the view field of radius of view, if there is barrier, current dimensions updating failure; If make d to certain one dimension of a certain particle _maxsecondary renewal is still unsuccessful, then the current dimensions updating failure of this particle, if make p to a certain particle _maxsecondary renewal is still unsuccessful, then this particle upgrades unsuccessfully; If all subgroups all upgrade failure, then termination of iterations proceed to step S7.

As prioritization scheme, step S66 is specially:

Judge whether to reach default evolution cycle R, if reach, first according to the death rate preset, from gene pool, Stochastic choice particle substitutes particle the most bad in each subgroup; Again according to adaptive value, if the adaptive value of a certain particle is less than the adaptive value of a certain particle in gene pool in each subgroup, then replace the corresponding particle in gene pool with this particle in each subgroup; Final updating population barycenter P _c.

As prioritization scheme, the method in step S7, robot being proceeded to next GOTO field is: if described current optimal movement orientation angle the local endpoint P of this Rolling Planning ^t' be positioned at current point P ^ty-axis increase to, then test is rotated counterclockwise to each GOTO field, otherwise dextrorotation transfer to test is done to each GOTO field.

Compared with prior art, the present invention has following beneficial effect:

(1) the present invention can realize robot and has static-obstacle thing and independent navigation under the working environment of Obstacle Position the unknown;

(2) the present invention introduces colony's barycenter strategy in the coevolution of many subgroups, improves the search capability of population;

(3) relative to the paths planning method of existing employing standard particle group algorithm (PSO) and coevolution particle cluster algorithm (CPSO), present invention employs survival of the fittest strategy, evolution convergence accuracy and runtime has more superiority;

(4) the present invention has to cook up in real time in complex barrier substance environment and effectively keeps away barrier path, has that robustness is good, solution efficiency advantages of higher.

Accompanying drawing explanation

Fig. 1 is overview flow chart of the present invention;

Fig. 2 is the process flow diagram of step S6 of the present invention;

Fig. 3 is the schematic diagram of working environment modeling in step S1 of the present invention;

Fig. 4 is the schematic diagram that in step S6 of the present invention, segmentation keeps away barrier;

Fig. 5 is that step S63 of the present invention falls into a trap and calculates the schematic diagram of segmented paths overall length;

Fig. 6 is the schematic diagram turning to domain test in step S7 of the present invention;

Fig. 7 is the path planning result in the embodiment of the present invention 1 Typical obstacles substance environment;

Fig. 8 is the comparison diagram that employing three kinds of algorithms of different carry out path planning;

Fig. 9 is the path planning result in the embodiment of the present invention 1 slit environment;

Figure 10 is the path planning result in the embodiment of the present invention 1 strip obstacle environment;

Figure 11 is the path planning result in the embodiment of the present invention 1 spill obstacle environment.

Embodiment

The present invention is described in detail by way of example below in conjunction with accompanying drawing.

Embodiment 1:

As shown in Figure 1, based on the robot path planning method of coevolution population rolling optimization, for planning the walking path of a robot, making some stationary obstruction that robot avoids in working environment, comprising the steps:

Step S1, carries out modeling to working environment, as shown in Figure 3, specifically comprises the steps:

Step S11, ignores the elevation information in working environment, sets up two-dimensional coordinate system.This step have ignored the elevation information of robot and barrier, the working environment of robot is described as the environment of a two-dimentional bounded.

Step S12, if the view field of robot is (r, θ), wherein r is radius of view, and θ is field-of-view angle; Obtain the starting point S of robot and the coordinate of impact point G; In two-dimensional coordinate system, describe barrier with convex polygon, the coordinate of barrier is unknown.

Step S13, the size according to robot is carried out expanded to barrier, and using robot as a particle process.In path planning of the present invention, ignore shape and the structure of robot itself, regarded as a particle and process.

Step S14, arranging current scrolling planning number of times t is 1, arranges the current point P of robot ^tfor starting point S, and by current point P ^tstored in terminal collection { P}.

Step S2, measuring robots current point P ^tand the distance between impact point G, if distance is less than default distance error ξ, then thinks and export terminal collection { P}, the acquisition programme path of each planning by arrival impact point; Otherwise proceed to step S3;

Step S3, according to robot current point P ^tcurrent optimal movement orientation angle is obtained with impact point G current optimal movement orientation angle refer to that current kinetic is towards the angle with two-dimensional coordinate system x-axis, current kinetic is oriented current point P ^tto the rectilinear direction of impact point G; Current optimal movement orientation angle computing formula as the formula (1):

(1)

Wherein, t is current scrolling planning number of times, if then make g _xfor described impact point G is at the coordinate figure of x-axis, G _yfor described impact point G is at the coordinate figure of y-axis, P _x ^tfor described current point P ^tat the coordinate figure of x-axis, P _y ^tfor described current point P ^tat the coordinate figure of y-axis.

Step S4, according to field-of-view angle θ, is divided into K=2 π/θ, according to current optimal movement orientation angle by the GOTO field of robot determine the GOTO field k of robot current kinetic towards place, current kinetic towards the GOTO field k at place computing formula as the formula (2):

(2)

Wherein, if k=0, then make k=K, wherein ceil () is the function that rounds up, and mod () is remainder function.

If step S5, clear in the view field of current robot, is that limit sets up local pole coordinate system with current point, according to the radius of view r preset and current optimal movement orientation angle obtain the local endpoint P of this planning ^t', local endpoint P ^t' computing formula as the formula (3):

(3)

Wherein, P ^t' _xfor described local endpoint P ^t' at the coordinate figure of x-axis, P ^t' _yfor described local endpoint P ^t' at the coordinate figure of y-axis, P _x ^tfor described current point P ^tat the coordinate figure of x-axis, P _y ^tfor described current point P ^tat the coordinate figure of y-axis;

By local endpoint P ^t' stored in terminal collection, { P}, makes t=t+1, and by local endpoint P ^t' as the current point P planned next time ^t, proceed to step S2; If have barrier in the view field of current robot, proceed to step S6.

As shown in Figure 2, step S6, adopts the mode of interative computation to carry out keeping away barrier process based on coevolution particle cluster algorithm; Specifically comprise the steps:

Step S61, initialization of population step, specifically comprises the steps:

Step S611, is divided into m section by radius of view r, obtains a segment radius λ, determines m segment movement orientation angle at random simultaneously , wherein, i=1,2 ... m, segment movement orientation angle be less than the field-of-view angle θ of robot; This step attempts attempting " around " barrier, as shown in Figure 4, according to segment radius λ and segment movement orientation angle with less segmentation and less angle m waypoint can be obtained in the view field of robot; Distance between adjacent two waypoints is segmented paths dist _i, thus obtain m segmented paths dist _i, by m segmented paths dist _iconnect to form the path overall length dist that robot avoiding obstacles within the scope of radius of view r need pass through _t.

Step S612, arranging some particles for carrying out the interative computation of particle cluster algorithm, being encoded to by particle by some segment movement orientation angles the sequence of composition, if the dimension of some particles is m, according to the GOTO field k of current kinetic towards place, defines the scope of some particles for [(k-1) θ, k θ], (k=1,2 ... K, K=2 π/θ).

Step S613, as shown in Figure 5, can determine path overall length dist according to the cosine law _t, path overall length dist _tcomputing formula as the formula (4):

${\mathrm{dist}}_t=\mathrm{\&lambda;}+\underset{i-2}{\overset{m}{\mathrm{\&Sigma;}}}{\mathrm{dist}}_i=r/m+\underset{i=2}{\overset{m}{\mathrm{\&Sigma;}}}\sqrt{{\left((i-1)\mathrm{\&lambda;}\right)}^2+{\left(\mathrm{i\&lambda;}\right)}^2+2\mathrm{i\&lambda;}(i-1)\mathrm{\&lambda;}\cos (x_i-x_{i-1})}$ (4)

Wherein, i=2,3 ..., m; λ=r/m, namely radius of view r makes the segment radius after m decile; First segmented paths dist ₁=λ, x _ifor the i-th dimension of particle after encoding;

Obtain local endpoint P ^t' to the air line distance of impact point G air line distance computing formula as the formula (5):

${\mathrm{dist}}_{P^{t\&prime;}G}=\sqrt{{({P^{t\&prime;}}_x-G_x)}^2+{\left({P^{t\&prime;}}_y{-G}_y\right)}^2}$ (5)

According to path overall length dist _tand air line distance set an adaptive value function, adaptive value function as the formula (6):

$\mathrm{Fit}={\mathrm{\&omega;}}_1{\mathrm{dist}}_t+{\mathrm{\&omega;}}_2{\mathrm{dist}}_{P^{t\&prime;}G}$ (6)

ω wherein ₁with ω ₂for weight coefficient.

Step S614, forms an initial population by some particles, carries out initialization to initial population; The current point P that initialization requirements is planned from this ^tstart, with segmented paths dist _ifor there is not barrier in the view field of radius of view, otherwise initialization failure; Step S614 comprises further:

Setting original dimensions is m, initial scale is N, before carrying out initialization, arranges the population scale of initial population much larger than initial scale N; The maximum initialization times p of particle is set _maxinitialization times d maximum with dimension _max; If certain one dimension of a certain particle is making d in initial population _maxstill unsuccessful after secondary initialization, then reinitialize current dimension; If this particle is through p _maxsecondary initialization is still failed, then think that current particle initialization is failed;

Abandon the particle of initialization failure, by the particle of successful initialization composition initial population; If the scale of initial population reaches initial scale N, then initialization success, proceeds to next step; Otherwise initialization failure, proceeds to step S7.

Step S615, if initialization success, is divided into s subgroup, according to the GOTO field k of current kinetic towards place, defines the scope [x of some particles by initial population _min, x _max] be [(k-1) θ, k θ]; The maximal rate arranging particle is 0.1 times of particle range.

Step S62, calculates the adaptive value of each particle in each subgroup, and sorts by adaptive value to the particle in each subgroup, then selects by the gene-ratio preset the particle that in each subgroup, adaptive value is higher, forms a gene pool.It is tactful that this step have employed " survival of the fittest ", and select " elite " particle optimum in each subgroup to create gene pool by certain gene-ratio, the particle in this gene pool is less " elite " particle of adaptive value.

Step S63, according to the adaptive value of particle each in each subgroup, obtains the optimum solution of each subgroup and try to achieve a population center of mass point P according to optimum solution _c; Population center of mass point P _cdefinition as the formula (7):

$P_c(R+1)=\frac{1}{s}\underset{1}{\overset{s}{\mathrm{\&Sigma;}}}{p}_{\mathrm{gd}}^i\left(R\right)i=\mathrm{1,2},...,s$ (7)

Wherein, R is evolution cycle, it is the globally optimal solution of i-th subgroup.Population center of mass point P _cthe i.e. mean value of each subgroup optimum solution, represent the desired center of whole population, it upgrades with evolution cycle, without speed component.Population center of mass point is the average of the globally optimal solution of each subgroup, it is a kind of desirable barycenter, a not actual particle, the reason introducing barycenter is, individual history optimum, subgroup global optimum and population will be followed desirable optimum when particle circles in the air in solution space, due to desirable more optimum than populations many in standard particle group evolutionary equation, so in theory particle flight time directivity is clearer and more definite, the randomness of particle flight can be reduced, there has also been between each subgroup certain " interchange ", therefore, it is possible to significantly improve the search capability of population.

Step S64, from then on step starts to carry out rolling interative computation, upgrades speed and the position of particle in each subgroup according to formula (8) and formula (9):

${v_i}^j(\mathrm{iter}+1)=\mathrm{\&omega;}\&times;{v_i}^j\left(\mathrm{iter}\right)+c_1{r}_1\&times;({P_i}^j\left(\mathrm{iter}\right)-{x_i}^j\left(\mathrm{iter}\right))+c_2{r}_2\&times;({g_i}^j\left(\mathrm{iter}\right)-{x_i}^j\left(\mathrm{iter}\right))+c_3{r}_3({P_c}^R-{x_i}^j\left(\mathrm{iter}\right))$

(8)

$x_i^j(\mathrm{iter}+1)=x_i^j\left(\mathrm{iter}\right)+v_i^j(\mathrm{iter}+1)$ (9)

Wherein iter is Evolution of Population algebraically, and j is subgroup mark, and i is particle label in subgroup, for the speed of i-th particle in a jth subgroup; P _i ^jfor the history optimum solution of i-th particle in a jth subgroup; for the optimum solution of subgroup, current particle place; c ₁and c ₂for Studying factors; c ₃for exchanging the factor; r ₁, r ₂, r ₃for the random number between [0,1]; ω is inertia weight, and inertia weight ω adopts linear decrease strategy, as the formula (10):

ω(iter)=ω _max-iter×(ω _max-ω _min)/iter _max(10)

The speed of particle and location boundary process are comprised: if the d of a certain particle i ties up speed in the j of subgroup $v_{i,j}^d>v_{\max },$ Then make $v_{i,j}^d=v_{\max };$ If $v_{i,j}^d<v_{\min },$ Then make $v_{i,j}^d=v_{\min };$ If $x_{i,j}^d>x_{\max },$ Then make $x_{i,j}^d=x_{\max };$ If $x_{i,j}^d<x_{\min },$ Then $x_{i,j}^d=x_{\min };$

Each dimension after particle position upgrades is required to include: from current point P ^tstart, with segmented paths dist _ifor there is not barrier in the view field of radius of view, if there is barrier, current dimensions updating failure; If make d to certain one dimension of a certain particle _maxsecondary renewal is still unsuccessful, then the current dimensions updating failure of this particle, if make p to a certain particle _maxsecondary renewal is still unsuccessful, then this particle upgrades unsuccessfully; If all subgroups all upgrade failure, then termination of iterations proceed to step S7; If be updated successfully, proceed to step S65.

Step S65, upgrades the globally optimal solution of the adaptive value of particle in each subgroup, the individual optimal solution of particle and subgroup.

Step S66, judges whether to reach default evolution cycle R, if do not reach evolution cycle R, directly proceeds to step S67; If reach evolution cycle R, first according to the death rate preset, from gene pool, Stochastic choice particle substitutes particle the most bad in each subgroup; If the adaptive value of a certain particle is less than the adaptive value of a certain particle in gene pool in each subgroup, then replace the corresponding particle in gene pool with this particle in each subgroup; Then Population Regeneration barycenter P _c; Finally proceed to step S67.It is tactful that this step have employed " survival of the fittest ", and the adaptive value of particle is less, then this particle is more outstanding.First eliminate according to death rate the particle that in each subgroup, a part of adaptive value is higher, from gene pool, select particle to replace; Eliminate the particle that in gene pool, adaptive value is higher again, from each subgroup, choose outstanding particle replace; Thus make the particle in gene pool and each subgroup all keep less adaptive value, make the speed of convergence of evolution faster, precision is higher.

Step S67, judges whether to reach default maximum iteration time iter _maxif reach maximum iteration time iter _max, then termination of iterations, obtains this local endpoint P planned according to the globally optimal solution of all subgroups and radius of view r ^t', and by local endpoint P ^t' add terminal collection and { in P}, make t=t+1, by local endpoint P ^t' as the current point P planned next time ^t, proceed to step S2; If do not reach maximum iteration time iter _max, then proceed to step S64 and continue iteration;

Step S7, carries out keeping away barrier failure handling, robot is proceeded to next GOTO field, and the method proceeding to next GOTO field is: if current optimal movement orientation angle the local endpoint P of this Rolling Planning ^t' be positioned at current point P ^ty-axis increase to, then test is rotated counterclockwise to each GOTO field, i.e. k=mod (k+i, K), i=1,2 ... K-1; Otherwise dextrorotation transfer to test is done to each GOTO field.When turning clockwise, k=mod (k-i, K) is pressed to k ~ 1 GOTO field and converts GOTO field, wherein i=0,1,2 ... k-1; K=mod (K-i, K) is pressed to K ~ k GOTO field and converts GOTO field, wherein i=0,1,2 ... K-k-1, as the formula (11):

After proceeding to next GOTO field, return step S61 and carry out keeping away barrier, as shown in Figure 6; Hinder unsuccessfully if robot all keeps away at all GOTO fields, then cannot realize keeping away barrier, unsuccessfully jump out.

In the present embodiment, following parameter is selected to be further described.

(1) population parameter: get initial scale N=20, subcluster number s=4, subgroup scale are 5, dimensionality of particle m=10, learning and exchange factor c ₁=c ₂=c ₃=2, inertia weight ω linearly drops to 0.4 from 0.9, weight coefficient ω in adaptive value function ₁=ω ₂=1, maximum iteration time iter _max=100; Evolution of Population cycle R=4, the maximum initialization times p of particle _maxinitialization times d maximum with dimension _maxfor p _max=d _max=20, gene-ratio is 0.2, and death rate is 0.2.

(2) robot running parameter: radius of view r=5, field-of-view angle is θ=π/3, and GOTO field is K=2 π/θ=6, distance error ξ=2.

Create the Typical obstacles thing working environment that size is 100 × 100, wherein there is several static-obstacle things, arranging starting point is S (10,10), and impact point is G (90,90), and the path cooked up as shown in Figure 7.

Systematic parameter is identical, select the point (45 closing on barrier, 29), standard particle group's algorithm (PSO), coevolution particle cluster algorithm (CPSO) and the present invention is used to adopt many subgroups coevolution particle cluster algorithm (MCPSO) of " survival of the fittest " strategy to carry out respectively keeping away to hinder testing, as shown in Figure 8, MCPSO algorithm has more superiority to the adaptive value change curve of three kinds of algorithms on evolution convergence accuracy and runtime as seen from the figure.

Under special obstacle environment, as the slit environment in Fig. 9, huge long strip type obstacle environment in Figure 10 and the matrix obstacle environment in Figure 11, adopt this programme also can effectively implement online path planning, Fig. 7, Fig. 8 and Fig. 9 are seen in the path cooked up.

Several specific embodiments of the application being only above, but the not limited thereto the changes that any person skilled in the art can think of of the application, all should drop in the protection domain of the application.

Claims (11)

1., based on the robot path planning method of coevolution population rolling optimization, for planning the walking path of a robot, making some stationary obstruction that robot avoids in working environment, it is characterized in that, comprise the steps:

Step S1, carries out modeling to described working environment;

Step S2, detects the distance between described robot current point and impact point, if described distance is less than default distance error, then thinks and exports the terminal collection of each planning by arrival impact point, obtain programme path; Otherwise proceed to step S3;

Step S3, obtains current optimal movement orientation angle according to described robot current point and impact point;

Step S4, according to the robot view field's angle preset, divides the GOTO field of robot; The GOTO field of motion towards place of current described robot is determined according to described current optimal movement orientation angle;

Step S5, if clear in the view field of current described robot, with described current point for limit sets up local pole coordinate system, the local endpoint of this planning is obtained according to the radius of view preset and described current optimal movement orientation angle, and by described local endpoint stored in described terminal collection, using described local endpoint as the current point planned next time, proceed to step S2; If have barrier in the view field of current described robot, proceed to step S6;

Step S6, adopts the mode of interative computation to carry out keeping away barrier process based on coevolution particle cluster algorithm; Described step S6 comprises further:

Step S61, is divided into several segmentations by described radius of view, determines several segment movement orientation angles simultaneously, and described segment movement orientation angle is less than described field-of-view angle; Some particles are set, each described particle are encoded to the sequence be made up of some described segment movement orientation angles, some described particles are formed an initial population, and initialization is carried out to described initial population, obtain an initial population; If initialization success, is divided into several subgroups by described initial population, proceeds to step S62; If initialization failure, proceeds to step S7;

Step S62, calculates the adaptive value of each particle in each described subgroup, and sorts by adaptive value to the particle in each described subgroup, then selects by the gene-ratio preset the particle that in each described subgroup, adaptive value is less, forms a gene pool;

Step S63, according to the adaptive value of each particle in each described subgroup, obtains the optimum solution of each described subgroup, and tries to achieve a population center of mass point according to described optimum solution;

Step S64, upgrades speed and the position of particle in each described subgroup; If be updated successfully, then proceed to step S65, if upgrade unsuccessfully, proceed to step S7;

Step S65, upgrades the globally optimal solution of the adaptive value of particle in each described subgroup, the individual optimal solution of particle and subgroup;

Step S66, judges whether to reach default evolution cycle, if do not reach described evolution cycle, directly proceeds to step S67; Otherwise upgrade described subgroup and described gene pool, then upgrade described population center of mass point, then proceed to step S67;

Step S67, judge whether to reach default maximum iteration time, if reached described maximum iteration time, then termination of iterations, obtain the local endpoint of this planning, and described local endpoint is added described terminal concentrate, using described local endpoint as the current point planned next time, proceed to step S2; If do not reach described maximum iteration time, then proceed to step S64 and continue iteration;

Step S7, carries out keeping away barrier failure handling, described robot is proceeded to next GOTO field, returns step S61 and carries out keeping away barrier; Hinder unsuccessfully if described robot all keeps away at all GOTO fields, then cannot realize keeping away barrier, unsuccessfully jump out.

2. the robot path planning method based on coevolution population rolling optimization according to claim 1, is characterized in that, the described working environment modeling of described step S1 specifically comprises the steps:

Step S11, ignores the elevation information in described working environment, sets up two-dimensional coordinate system;

Step S12, if the view field of described robot is (r, θ), wherein r is radius of view, and θ is field-of-view angle; Obtain the starting point S of described robot and the coordinate of impact point G; Describe described barrier with convex polygon, the coordinate of described barrier is unknown;

Step S13, the size according to described robot is carried out expanded to described barrier, and using described robot as a particle process;

Step S14, arranging current scrolling planning number of times t is 1, arranges the current point P of described robot ^tfor described starting point S, and by described current point P ^tstored in terminal collection { P}.

3. the robot path planning method based on coevolution population rolling optimization according to claim 2, is characterized in that, described step S3 is specially:

Described current optimal movement orientation angle refer to that current kinetic is towards the angle with described two-dimensional coordinate system x-axis, described current kinetic is oriented described current point P ^tto the rectilinear direction of described impact point G; Described current optimal movement orientation angle computing formula such as formula shown in (1):

Wherein, t is current scrolling planning number of times, if then make g _xfor described impact point G is at the coordinate figure of x-axis, G _yfor described impact point G is at the coordinate figure of y-axis, P _x ^tfor described current point P ^tat the coordinate figure of x-axis, P _y ^tfor described current point P ^tat the coordinate figure of y-axis.

4. the robot path planning method based on coevolution population rolling optimization according to claim 3, is characterized in that, described step S4 is specially:

According to described field-of-view angle θ, the GOTO field of described robot is divided into K=2 π/θ, according to described current optimal movement orientation angle determine the GOTO field k of described robot current kinetic towards place, described current kinetic towards the computing formula of the GOTO field k at place such as formula shown in (2):

Wherein, if k=0, then make k=K, wherein ceil () is the function that rounds up, and mod () is remainder function.

5. the robot path planning method based on coevolution population rolling optimization according to claim 4, is characterized in that, described step S5 is specially:

Described local endpoint P ^t'computing formula such as formula shown in (3):

Wherein, P ^t' _xfor described local endpoint P ^t'at the coordinate figure of x-axis, P ^t' _yfor described local endpoint P ^t'at the coordinate figure of y-axis, P _x ^tfor described current point P ^tat the coordinate figure of x-axis, P _y ^tfor described current point P ^tat the coordinate figure of y-axis; By described local endpoint P ^t'{ after P}, t=t+1 is made stored in described terminal collection.

6. the robot path planning method based on coevolution population rolling optimization according to claim 5, is characterized in that, described step S61 comprises the steps: further

Step S611, is divided into m section by described radius of view r, obtains a segment radius λ, determines m described segment movement orientation angle at random wherein, obtain some segmented paths dist _i, by described some segmented paths dist _iconnect to form described robot within the scope of described radius of view r, avoid the path overall length dist that described barrier need pass through _t;

Step S612, arranges some particles for carrying out the interative computation of particle cluster algorithm, is encoded to by described particle by some described segment movement orientation angles the sequence of composition, if the dimension of described some particles is m, according to the GOTO field k of described current kinetic towards place, defines the scope of described some particles for [(k-1) θ, k θ], wherein, k=1,2 ... K, K=2 π/θ;

Step S613, determines described path overall length dist _t, described path overall length dist _tcomputing formula such as formula shown in (4):

Wherein, i=2,3 ..., m; λ=r/m, namely radius of view r makes the section length after m decile; x _ifor the i-th dimension of particle after encoding;

Obtain described local endpoint P ^t'to the air line distance of described impact point G described air line distance computing formula such as formula shown in (5):

According to described path overall length dist _twith described air line distance set an adaptive value function, described adaptive value function is such as formula shown in (6):

ω wherein ₁with ω ₂for weight coefficient;

Step S614, forms an initial population by described some particles, carries out initialization to initial population; The current point P that described initialization requirements is planned from this ^tstart, with described segmented paths dist _ifor there is not barrier in the view field of radius of view, otherwise initialization failure;

Step S615, if initialization success, is divided into s subgroup, according to the GOTO field k of described current kinetic towards place, defines the scope [x of described some particles by described initial population _min, x _max] be [(k-1) θ, k θ]; The maximal rate arranging particle is 0.1 times of particle range.

7. the robot path planning method based on coevolution population rolling optimization according to claim 6, is characterized in that, the initialization in described step S614 is specially:

Setting original dimensions is m, initial scale is N, before carrying out initialization, arranges the population scale of described initial population much larger than initial scale N; The maximum initialization times p of particle is set _maxinitialization times d maximum with dimension _max; If certain one dimension of a certain particle is making d in described initial population _maxstill unsuccessful after secondary initialization, then reinitialize current dimension; If this particle is through p _maxsecondary initialization is still failed, then think that current particle initialization is failed;

Abandon the particle of initialization failure, the particle of successful initialization is formed described initial population; If the scale of described initial population reaches initial scale N, then initialization success; Otherwise, initialization failure.

8. the robot path planning method based on coevolution population rolling optimization according to claim 7, is characterized in that, population center of mass point P described in described step S63 _cdefinition such as formula shown in (7):

Wherein, R is evolution cycle, it is the globally optimal solution of i-th subgroup.

9. the robot path planning method based on coevolution population rolling optimization according to claim 8, is characterized in that, described step S64 is specially:

Speed and the position of particle in each described subgroup is upgraded according to formula (8) and formula (9):

Wherein iter is Evolution of Population algebraically, and j is subgroup mark, and i is particle label in subgroup, for the speed of i-th particle in a jth subgroup; P _i ^jfor the history optimum solution of i-th particle in a jth subgroup; for the optimum solution of subgroup, current particle place; c ₁and c ₂for Studying factors; c ₃for exchanging the factor; r ₁, r ₂, r ₃for the random number between [0,1]; ω is inertia weight, and described inertia weight ω adopts linear decrease strategy, shown in (10):

ω(iter)＝ω _max-iter×(ω _max-ω _min)/iter _max(10)

The speed of particle and location boundary process are comprised: if the d of a certain particle i ties up speed in the j of subgroup then make if then make if then make if then

Each dimension after particle position upgrades is required to include: from described current point P ^tstart, with described segmented paths dist _ifor there is not barrier in the view field of radius of view, if there is barrier, current dimensions updating failure; If make d to certain one dimension of a certain particle _maxsecondary renewal is still unsuccessful, then the current dimensions updating failure of this particle, if make p to a certain particle _maxsecondary renewal is still unsuccessful, then this particle upgrades unsuccessfully; If all subgroups all upgrade failure, then termination of iterations proceed to step S7.

10. the robot path planning method based on coevolution population rolling optimization according to claim 9, is characterized in that, described step S66 is specially:

Judge whether to reach default evolution cycle R, if reach, first according to the death rate preset, from described gene pool, Stochastic choice particle substitutes particle the most bad in each described subgroup; If the adaptive value of a certain particle is less than the adaptive value of a certain particle in described gene pool in each described subgroup, then replace the corresponding particle in described gene pool with this particle in each described subgroup; Final updating population barycenter P _c.

11. robot path planning methods based on coevolution population rolling optimization according to claim 10, is characterized in that, the method in described step S7, described robot being proceeded to next GOTO field is: if described current optimal movement orientation angle the local endpoint P of this Rolling Planning ^t'be positioned at current point P ^ty-axis forward, then test is rotated counterclockwise to each GOTO field, otherwise dextrorotation transfer to test is done to each GOTO field.