CN102831318A - Task allocation algorithm based on individual capacity in heterogeneous multi-robot system - Google Patents

Abstract

The invention belongs to the field of research of robots, and the focus is of a model for research of individual capacity of each robot in a heterogeneous multi-robot system. The capacity of the individual robots in the multi-robot system is firstly divided into three types, the first type is time-invariant A type physical capacity, the second type is B type capacity which is weakened along with time, and the third type is of C type capacity related to individual willingness factor of each robot. The three types of the capacity are respectively evaluated, a capacity evaluation model and method for the individual robots is finally formed, factors affecting each type of capacity, in particular to the C type capacity with relatively complex changes, is discussed, and influence of parameter changes on the C type capacity is further obtained by a contrast test. The capacity model can solve the problem that the capacity of each robot in the heterogeneous multi-robot system is complex, changeable and difficult to evaluate, the capacity of the individual robots can be simultaneously evaluated in a real-time manner, a task list is generated on the basis of a task model, and a task allocation scheme is proposed based on the task list.

Description

In the heterogeneous multi-robot system based on the task allocation algorithms of individual capability

Technical field

The invention belongs to the robot research field, proposed a kind of method for solving that is applicable to the robot capability tabulation of analyzing based on the robot individual difference of heterogeneous multi-robot system, and provided task allocation algorithms based on this capabilities list.

Background technology

In order to let multi-robot system develop towards the direction of people's expectation, R.A.Brooks has proposed behavioristic thought; No longer will formulate meticulous and complicated control algolithm as making great efforts target for robot system; But the actual conditions in simulating nature circle; Attempting is that robot is equipped with different basic acts, revests the ability of its combination basic act, thereby realizes advanced ability.On the basis of behavior doctrine thought, the research of multi-robot system has obtained very big development.Yet it is extremely complicated that the inner relation of multi-robot system remains, if deal with improperly, not only can not bring into play the function of expection, probably also makes individual machine people's performance also be affected.

Take a broad view organic sphere, colourful, vigorous.Its topmost characteristic is a diversity, and promptly individual difference that is to say to exist varied rather than single simple population.In like manner, a colony inside, personal feature also is very obviously to enrich.This expects us, and in the multi-robot system that produces in the simulation biosystem, individual difference also should give abundant attention, and this is likely that system is able to the basis of constantly improving and developing.Therefore; We consider individual character is incorporated in the multi-robot system, give different robots different individual character key elements, and allow them to pass through the assessment of work and the trial summing up experience in environment; Constantly adjust the individual key element that self had; Reaching continuous adjustment self strategy better conforming and the requirement of work buddies characteristic, match according to needs and other individualities of task, thereby more bring into play the potentiality of self and entire system.

Task Distribution is one of key problem in the multi-robot system research, and scholars have done the research of Task Assignment Model respectively from many aspects.Many scholars also obtain inspiration and regard the individual machine people among the MRS as in the biotic population a object from the behavior of natural biology; Study so introduced the principle of sociology and science of economic management, proposed method for allocating tasks based on market economy etc. like Gerkey.If each robot emulatively accomplishes a task, in order effectively to handle this competition conflict,, general criterion comparatively fair with regard to needing assessed the ability of robot, thereby the task that makes can more efficient, high-quality completion.

Summary of the invention

The objective of the invention is to solve the individual capability evaluation problem of robot in the isomery robot system, concrete grammar is following:

(1) the isomery robot capability is described

Generally speaking, be divided into three kinds to the ability of robot: the category-A ability is the inherent base attribute that can not change, and comprising: the sense of hearing, vision and motion, communication bandwidth; The category-B ability is the ability of successively decreasing in time and consuming, or perhaps the robot energy that can supply consume, and comprising: computing velocity, electric power deposit and mechanical property; C class ability is other abilities outside A, the category-B ability, is made up of at random friendly degree and the emotional factors that personalize such as cooperation wish, robustness, embodies the subjective ability to work of robot.That ability attribute in every type of ability is divided into again is excellent, good, in, differ from four ranks, the power of expression ability.Every type of ability can continue interpolation ability attribute according to actual conditions.

For every type of basic capacity, it is following to define its competence set:

S={s _i|i＝1,2,...,N}

Wherein N representes the number of ability attribute in every type of ability.Use P _iThe weight of representing i kind ability, contribution margin in other words.

Then the probability space model of robot capability can be expressed as

$\left(\begin{array}{c}S\\ P\end{array}\right)=\left(\begin{array}{cccc}{s}_1& s_2& ...& s_N\\ P_1& P_2& ...& P_N\end{array}\right)$

(2) robot individual capability assessment

The ability of robot can be expressed as:

ΔE=(αE ₁+βE ₂)(α+β)/(α+β+γ)+γΨ

=(αE ₁+βE ₂)(α+β)+γΨ

Wherein, E ₁Be inherent basic hardware ability, the category-A ability that can not change along with change of time; E ₂Be the category-B ability of successively decreasing along with the time, Ψ representes the matching degree between the demand weight matrix of robot C class ability.Different parameters obtains the robot of different characteristics of personality.Suppose three parameter normalization, that is:

α+β+γ=1

This system of parameters base is in the difference of task and difference.Each task is come interim, according to the weight to each item ability need of task, confirms three parameters, calculates the ability of each robot in the multi-robot system then, and the higher robot of selective power executes the task according to qualifications.That will explain every type of ability below specifically finds the solution thinking.

1) category-A capability evaluation

For the category-A ability, its expression is fairly simple

$E_i=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{P}_i\×{\mathrm{\φ}}_i$

P wherein _iUnder current task, the weight of attribute ability i in the category-A ability, φ _iBe the grade of attribute ability i, be divided into and differ from four grades in good that respectively corresponding 1,3/4,1/2,0 four class value thinks promptly when robot does not possess this ability that its ability grade is for poor.

2) category-B capability evaluation

The B ability is along with the ability value of time decay, is not having under the situation of load, promptly is not assigned with under the situation of task, and its ability presents the nature attenuation function.

If

is individual capability i initial ability value, then t+1 individual capability constantly can be expressed as

φ ^(t+1)=φ ^(t)·A

Wherein A is called transition matrix, also can be called damping matrix simultaneously, means that As time goes on the ability of robot slowly reduces, and promptly energy is consumed.

Under loaded situation, its transition matrix becomes B, and B also is an attenuation function

|B|<|A|<1

The attenuation degree that is it is bigger than A.

This moment, individual capability can be expressed as:

φ ^(t+1)=φ ^(t)·B

The category-B ability can be expressed as

$E_2=\underset{i=0}{\overset{N}{\mathrm{\&Sigma;}}}{P}_i\&times;{\mathrm{\&phi;}}_i^{\left(t\right)}$

3) C class capability evaluation

C class ability is the ability that best embodies the individual wish of robot, through the incompatible difference that forms individuality of the random groups of each ability attribute.Wherein weight is satisfied in other words for the probability of each ability attribute

0≤P wherein _i≤1 (i=1,2 ... N)

At any time, the probability sum of each ability all is 1, and this each capability state of explanation is a mutual restriction, and a kind of increase of competency degree will inevitably make other capability state intensity reduce.

For each robot, at task arriving moment t, a C class ability weight is arranged all, represent as follows:

${\stackrel{\&OverBar;}{P}}^t=[p_1^t,p_2^t,...,p_N^t]$

And for the task i of each decomposition, a C class ability need weight is arranged all, expression as follows:

${\stackrel{\&OverBar;}{P}}_i=[p_1^i,p_2^i,...,p_N^i]$

The matching degree that defines two weight matrixes is following:

$\mathrm{\&psi;}=\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}\frac{p_k^t}{p_k^i}$

A. spontaneous transfer individual capability Research on differences

On the basis of probability space, regard the change procedure of ability as a stochastic process, and come the spontaneous transfer process of descriptive power state with Markov chain.And always suppose and shift that promptly under the situation that does not have the task load, the transfer of ability trends towards the state that each ability attribute probability is evenly distributed to calmness point, promptly probability is 1/N.The ability attribute that is equivalent to not highlight, promptly mediocre state.

At this moment ${\stackrel{\&OverBar;}{P}}^t=\left(\begin{array}{cccc}{\stackrel{\&OverBar;}{P}}_1^t& {\stackrel{\&OverBar;}{P}}_2^t& ...& {\stackrel{\&OverBar;}{P}}_N^t\end{array}\right)$ Being illustrated in the probability distribution of the ability of t constantly, is the initial ability probability distribution during t=0.

${\stackrel{\&OverBar;}{P}}^{t+1}={\stackrel{\&OverBar;}{P}}^t\&CenterDot;A$

Wherein

${\stackrel{\&OverBar;}{P}}^t=[p_1^t,p_2^t,...,p_N^t]$

$A={\left\{a_{\mathrm{ij}}\right\}}_{\mathrm{MxN}}=\left[\begin{array}{cccc}\frac{\mathrm{\&theta;}{\mathrm{\&pi;}}_1^{*}-(N-1)}{\mathrm{\&theta;}{\mathrm{\&pi;}}_1^{*}}& \frac{1}{\mathrm{\&theta;}{\mathrm{\&pi;}}_1^{*}}& ...& \frac{1}{\mathrm{\&theta;}{\mathrm{\&pi;}}_1^{*}}\\ \frac{1}{\mathrm{\&theta;}{\mathrm{\&pi;}}_2^{*}}& \frac{\mathrm{\&theta;}{\mathrm{\&pi;}}_2^{*}-(N-1)}{\mathrm{\&theta;}{\mathrm{\&pi;}}_2^{*}}& ...& \frac{1}{\mathrm{\&theta;}{\mathrm{\&pi;}}_2^{*}}\\ ...& ...& ...& ...\\ \frac{1}{\mathrm{\&theta;}{\mathrm{\&pi;}}_N^{*}}& \frac{1}{\mathrm{\&theta;}{\mathrm{\&pi;}}_N^{*}}& ...& \frac{\mathrm{\&theta;}{\mathrm{\&pi;}}_N^{*}-(N-1)}{\mathrm{\&theta;}{\mathrm{\&pi;}}_N^{*}}\end{array}\right]$

θ is a undetermined parameter, ${\stackrel{\&OverBar;}{\mathrm{\&pi;}}}^{*}=\left[\begin{array}{cccc}{\mathrm{\&pi;}}_1^{*}& {\mathrm{\&pi;}}_2^{*}& ...& {\mathrm{\&pi;}}_N^{*}\end{array}\right]$ Be limiting probabilities.

The spontaneous transfer process of capability state; Promptly individual ability shifts difference; Depend on matrix A; Therefore, parameter θ and

confirm, are the keys to individual difference research.

B. stimulate and shift the individual capability Research on differences

When robot when t receives an assignment constantly, the ability that occurs under the spread effect is stimulated transfer process, this moment To stimulate the starting point of transfer process as ability, be called the initial probability distribution that ability shifts, with π=[π ₁π ₂... π _N] expression,

At this moment represent this ability transfer process with HMM, with setting up the five-tuple model

$\mathrm{\&lambda;}=(N,M,{\widehat{\stackrel{\&OverBar;}{P}}}^0,\hat{A},\hat{B})$

Wherein N is the ability dimension; It is number; M stimulates kind; Be defaulted as M=N;

is initial ability probability distribution vector;

is state transition probability matrix, and observation matrix is

$\hat{B}=\left[\begin{array}{c}{B}_{1\&CenterDot;}\\ B_{2\&CenterDot;}\\ ...\\ B_{N\&CenterDot;}\end{array}\right]=\left[\begin{array}{cccc}a& b& ...& b\\ b& a& ...& b\\ ...& ...& ...& ...\\ b& b& ...& a\end{array}\right]$

$a=\frac{r}{N-1+r}$

$b=\frac{1}{N-1+r}$

R>1, r is an adjustable parameter.

Five-tuple according to the HMM model is described; Learn in the capability state transfer process; Individual capacity variance; Depend on matrix

and

therefore; Parameter r, θ and

confirm, are the keys to individual difference research.

(2) the isomery robot task is distributed

When task is come temporarily, for each branch task, the isomery robot that is in the relevant work zone can generate a capabilities list.For task, can generate a capabilities list, this capabilities list comprises: task, robot, the ability of robot under this task, and task rank.Accordingly, for each robot, can both generate a capabilities list, this capabilities list comprises the ability under robot, task, this task condition and whether load is arranged.According to the order of task priority, according to tab sequential, allocating task is promptly chosen in the capabilities list of this task and is come top robot, generally speaking, arranges task according to task priority, for the distribution of the identical a plurality of tasks of priority.

With<e _i, R _j, n, level _i, assigned _i>Expression task list, wherein e _iRepresent the i subtasks, according to level arrangement, the task series arrangement of same level, n representes the capabilities list order of the robot that this task is corresponding, if n=1, then corresponding R _iExpression is for task e _iRobot with highest-capacity; The ability that the corresponding robot of the big more expression of n carries out this task is more little, and level representes the rank of task, and assigned=true representes that task distributes; Assigned=false representes that task is also unallocated, and assigned=0 representes undetermined;<r ₁, e _k, m, load ₁>Expression individual machine people's capabilities list, R ₁Represent l robot, m representes the capabilities list order of the subtask that robot is corresponding, if m=1 then carries out corresponding e _kDuring task, the R of robot ₁Ability is the strongest, and the ability that the big more expression of m robot carries out corresponding task is more little, load ₁The expression R of robot ₁Whether be assigned with task.

Program circuit is following:

1) sets original state<e _i, R _j, n, level _i, assigned _i>, i=1, level=A;

2) make n=1, assigned _i=false, if<e _i, R _j, n, level _i, assigned _i>Be sky, then forward 7 to);

3) read<e _i, R _j, n, level _i, assigned _i>The corresponding R of robot _j, draw the value of j;

4) for<r ₁, e _k, m, load ₁>, making l=j, m=1 reads corresponding task e _k, obtain the value of k;

5) if load ₁=false, and k=i are then task e _iDistribute to the R of robot _j, make assigned _i=true, load ₁=true; If k is not equal to i, load ₁=false reads<e _k, R _j, n, level _k>, obtain level _k, if level _k=! Level _i, then e _iDistribute to the R of robot _j, make assigned _i=true, load ₁=true; Otherwise, make assigned _i=0; If load ₁=true makes n=n+1 turn to step 3);

6) judge level _I+1=level _i, if, then make p=i, i=i+1 turns to step 2), if not, then read<e _I1, R _j, n, level _I-1, 0>If, be empty, then make i=p+1, turn to step 2), otherwise, then make i=il, execution in step 3) 4) 5)

7) termination routine

The present invention has set up the individual capability model of a kind of robot, this model can the analog physical world in the transfer process of robot capability, the ability that the real-time assessment robot is individual is for the Task Distribution of robot system lays the first stone.

Embodiment

Suppose following scene: a task is broken down into four subtasks: e ₁, e ₂, e ₃, e ₄, in operative scenario, have four R of robot _A, R _B, R _C, R _D, all there is the right of each task of execution in each robot, and environmental factor is provided with as follows:

Table 1 task environment is described

The capabilities list of corresponding task is as shown in the table:

The capabilities list of table 2 task

Can know subtask e from top tabulation ₁The highest rank is just arranged, subtask e ₁, e ₂Has identical rank.

Simultaneously can obtain the R of robot _ACorresponding capabilities list is as shown in the table, from table, can find out subtask e ₁Corresponding ability is the highest, and because e ₁Have the highest rank, this just means has the subtask of highest level e ₁Should distribute to the R of robot _A

The R of table 3 robot _ACapabilities list

From the R of robot _BCapabilities list in can find out R _BThe e in the subtask ₃Have the highest ability under the condition, but subtask e ₂Under the condition, highest-capacity also belongs to R _B, last, because subtask e ₂, e ₃Have identical rank, and R _BAt task e ₃Has higher ability down, so task e ₃Be assigned to R _B

The R of table 4 robot _BCapabilities list

Because e ₃Be assigned to the R of robot _B, and the R of robot _CBe task e ₂Optimal selection, at the R of robot _CCapabilities list in, at task e ₄Under the condition, the R of robot _CThe highest ability is arranged, but task e ₂Compare e ₄Rank high, so the R of robot _CShould be assigned with task e ₂.

The R of table 5 robot _CCapabilities list

At last, other task of lowermost level e ₄Be assigned to the R of robot _DThe method for allocating tasks that the utilization front is introduced, last allocation result is as follows:

Table 6 Task Distribution result

This example has comprised three kinds of situation mentioning in the top method for allocating tasks; This example has proved that above-mentioned method for allocating tasks based on capabilities list can guarantee that the robot selection with highest-capacity has the task of highest level simultaneously, thereby improves Task Distribution efficient.

Claims (5)

1. robot capability is described and assessment in the heterogeneous multi-robot system, and is specific as follows:

(1) the isomery robot capability is described

Generally speaking, be divided into three kinds to the ability of robot: the category-A ability is the inherent base attribute that can not change, and comprising: the sense of hearing, vision and motion, communication bandwidth; The category-B ability is the ability of successively decreasing in time and consuming, or perhaps the robot energy that can supply consume, and comprising: computing velocity, electric power deposit and mechanical property; C class ability is other abilities outside A, the category-B ability, is made up of at random friendly degree and the emotional factors that personalize such as cooperation wish, robustness, embodies the subjective ability to work of robot; That ability attribute in every type of ability is divided into again is excellent, good, in, differ from four ranks, the power of expression ability; Every type of ability can continue interpolation ability attribute according to actual conditions;

For every type of basic capacity, it is following to define its competence set:

S={s _i|i＝1,2,...,N}

Wherein N representes the number of ability attribute in every type of ability; Use P _iThe weight of representing i kind ability, contribution margin in other words;

Then the probability space model of robot capability can be expressed as

$\left(\begin{array}{c}S\\ P\end{array}\right)=\left(\begin{array}{cccc}{s}_1& s_2& ...& s_N\\ P_1& P_2& ...& P_N\end{array}\right)$

(2) robot individual capability assessment

The ability of robot can be expressed as:

ΔE=(αE ₁+βE ₂)(α+β)/(α+β+γ)+γΨ

=(αE ₁+βE ₂)(α+β)+γΨ

Wherein, E ₁Be inherent basic hardware ability, the category-A ability that can not change along with change of time; E ₂Be the category-B ability of successively decreasing along with the time, Ψ representes the matching degree between the demand weight matrix of robot C class ability; Different parameters obtains the robot of different characteristics of personality; Suppose three parameter normalization, that is:

α+β+γ=1

This system of parameters is several different according to the difference of task; Each task is come interim, according to the weight to each item ability need of task, confirms three parameters, calculates the ability of each robot in the multi-robot system then, and the higher robot of selective power executes the task according to qualifications.

2. algorithm according to claim 1 is characterized in that, said category-A capability evaluation is specific as follows:

For the category-A ability, its expression is fairly simple

$E_i=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{P}_i\&times;{\mathrm{\&phi;}}_i$

P wherein _iUnder current task, the weight of attribute ability i in the category-A ability, φ _iBe the grade of attribute ability i, be divided into and differ from four grades in good that respectively corresponding 1,3/4,1/2,0 four class value thinks promptly when robot does not possess this ability that its ability grade is for poor.

3. algorithm according to claim 1 is characterized in that, said category-B capability evaluation is specific as follows:

The B ability is along with the ability value of time decay, is not having under the situation of load, promptly is not assigned with under the situation of task, and its ability presents the nature attenuation function;

If is individual capability i initial ability value, then t+1 individual capability constantly can be expressed as

φ ^(t+1)=φ ^(t)·A

Wherein A is called transition matrix, also can be called damping matrix simultaneously, means that As time goes on the ability of robot slowly reduces, and promptly energy is consumed;

Under loaded situation, its transition matrix becomes B, and B also is an attenuation function

|B|<|A|<1

The attenuation degree that is it is bigger than A;

This moment, individual capability can be expressed as:

φ ^(t+1)=φ ^(t)·B

The category-B ability can be expressed as

$E_2=\underset{i=0}{\overset{N}{\mathrm{\&Sigma;}}}{P}_i\&times;{\mathrm{\&phi;}}_i^{\left(t\right)}.$

4. algorithm according to claim 1 is characterized in that, said C class capability evaluation is specific as follows:

C class ability is the ability that best embodies the individual wish of robot, through the incompatible difference that forms individuality of the random groups of each ability attribute; Wherein weight is satisfied in other words for the probability of each ability attribute 0≤P wherein _i≤1 (i=1,2 ... N)

At any time, the probability sum of each ability all is 1, and this each capability state of explanation is a mutual restriction, and a kind of increase of competency degree will inevitably make other capability state intensity reduce;

For each robot, at task arriving moment t, a C class ability weight is arranged all, represent as follows:

${\stackrel{\&OverBar;}{P}}^t=[p_1^t,p_2^t,...,p_N^t]$

And for the task i of each decomposition, a C class ability need weight is arranged all, expression as follows:

${\stackrel{\&OverBar;}{P}}_i=[p_1^i,p_2^i,...,p_N^i]$

The matching degree that defines two weight matrixes is following:

$\mathrm{\&Psi;}=\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}\frac{p_k^t}{p_k^i}$

A. spontaneous transfer individual capability Research on differences

On the basis of probability space, regard the change procedure of ability as a stochastic process, and come the spontaneous transfer process of descriptive power state with Markov chain; And always suppose and shift that promptly under the situation that does not have the task load, the transfer of ability trends towards the state that each ability attribute probability is evenly distributed to calmness point, promptly probability is 1/N; The ability attribute that is equivalent to not highlight, promptly mediocre state;

At this moment ${\stackrel{\&OverBar;}{P}}^t=\left(\begin{array}{cccc}{\stackrel{\&OverBar;}{P}}_1^t& {\stackrel{\&OverBar;}{P}}_2^t& ...& {\stackrel{\&OverBar;}{P}}_N^t\end{array}\right)$ Being illustrated in the probability distribution of the ability of t constantly, is the initial ability probability distribution during t=0;

${\stackrel{\&OverBar;}{P}}^{t+1}={\stackrel{\&OverBar;}{P}}^t\&CenterDot;A$

Wherein

${\stackrel{\&OverBar;}{P}}^t=[p_1^t,p_2^t,...,p_N^t]$

$A={\left\{a_{\mathrm{ij}}\right\}}_{\mathrm{MxN}}=\left[\begin{array}{cccc}\frac{\mathrm{\&theta;}{\mathrm{\&pi;}}_1^{*}-(N-1)}{\mathrm{\&theta;}{\mathrm{\&pi;}}_1^{*}}& \frac{1}{\mathrm{\&theta;}{\mathrm{\&pi;}}_1^{*}}& ...& \frac{1}{\mathrm{\&theta;}{\mathrm{\&pi;}}_1^{*}}\\ \frac{1}{\mathrm{\&theta;}{\mathrm{\&pi;}}_2^{*}}& \frac{\mathrm{\&theta;}{\mathrm{\&pi;}}_2^{*}-(N-1)}{\mathrm{\&theta;}{\mathrm{\&pi;}}_2^{*}}& ...& \frac{1}{\mathrm{\&theta;}{\mathrm{\&pi;}}_2^{*}}\\ ...& ...& ...& ...\\ \frac{1}{\mathrm{\&theta;}{\mathrm{\&pi;}}_N^{*}}& \frac{1}{\mathrm{\&theta;}{\mathrm{\&pi;}}_N^{*}}& ...& \frac{\mathrm{\&theta;}{\mathrm{\&pi;}}_N^{*}-(N-1)}{\mathrm{\&theta;}{\mathrm{\&pi;}}_N^{*}}\end{array}\right]$

θ is a undetermined parameter, ${\stackrel{\&OverBar;}{\mathrm{\&pi;}}}^{*}=\left[\begin{array}{cccc}{\mathrm{\&pi;}}_1^{*}& {\mathrm{\&pi;}}_2^{*}& ...& {\mathrm{\&pi;}}_N^{*}\end{array}\right]$ Be limiting probabilities;

The spontaneous transfer process of capability state; Promptly individual ability shifts difference; Depend on matrix A; Therefore, parameter θ and confirm, are the keys to individual difference research;

B. stimulate and shift the individual capability Research on differences

When robot when t receives an assignment constantly, the ability that occurs under the spread effect is stimulated transfer process, this moment

To stimulate the starting point of transfer process as ability, be called the initial probability distribution that ability shifts, with π=[π ₁π ₂... π _N] expression,

At this moment represent this ability transfer process with HMM, with setting up the five-tuple model

$\mathrm{\&lambda;}=(N,M,{\widehat{\stackrel{\&OverBar;}{P}}}^0,\hat{A},\hat{B})$

Wherein N is the ability dimension; It is number; M stimulates kind; Be defaulted as M=N;

is initial ability probability distribution vector; is state transition probability matrix, and observation matrix is

$\hat{B}=\left[\begin{array}{c}{B}_{1\&CenterDot;}\\ B_{2\&CenterDot;}\\ ...\\ B_{N\&CenterDot;}\end{array}\right]=\left[\begin{array}{cccc}a& b& ...& b\\ b& a& ...& b\\ ...& ...& ...& ...\\ b& b& ...& a\end{array}\right]$

$a=\frac{r}{N-1+r}$

$b=\frac{1}{N-1+r}$

R>1, r is an adjustable parameter;

Five-tuple according to the HMM model is described; Learn in the capability state transfer process; Individual capacity variance; Depend on matrix

and

therefore; Parameter r, θ and

confirm, are the keys to individual difference research.

5. algorithm according to claim 1 is characterized in that, said isomery robot task is distributed specific as follows:

When task is come temporarily, for each branch task, the isomery robot that is in the relevant work zone can generate a capabilities list; For task, can generate a capabilities list, this capabilities list comprises: task, robot, the ability of robot under this task, and task rank; Accordingly, for each robot, can both generate a capabilities list, this capabilities list comprises the ability under robot, task, this task condition and whether load is arranged; According to the order of task priority, according to tab sequential, allocating task is promptly chosen in the capabilities list of this task and is come top robot, generally speaking, arranges task according to task priority, for the distribution of the identical a plurality of tasks of priority;

With<e _i, R _j, n, level _i, assigned _i>Expression task list, wherein e _iRepresent the i subtasks, according to level arrangement, the task series arrangement of same level, n representes the capabilities list order of the robot that this task is corresponding, if n=1, then corresponding R _iExpression is for task e _iRobot with highest-capacity; The ability that the corresponding robot of the big more expression of n carries out this task is more little, and level representes the rank of task, and assigned=true representes that task distributes; Assigned=false representes that task is also unallocated, and assigned=0 representes undetermined;<r ₁, e _k, m, load ₁>Expression individual machine people's capabilities list, R ₁Represent l robot, m representes the capabilities list order of the subtask that robot is corresponding, if m=1 then carries out corresponding e _kDuring task, the R of robot ₁Ability is the strongest, and the ability that the big more expression of m robot carries out corresponding task is more little, load ₁The expression R of robot ₁Whether be assigned with task;

Program circuit is following:

1) sets original state<e _i, R _j, n, level _i, assigned _i>, i=1, level=A;

2) make n=1, assigned _i=false, if<e _i, R _j, n, level _i, assigned _i>Be sky, then forward 7 to);

3) read<e _i, R _j, n, level _i, assigned _i>The corresponding R of robot _j, draw the value of j;

4) for<r ₁, e _k, m, load ₁>, making l=j, m=1 reads corresponding task e _k, obtain the value of k;

5) if load ₁=false, and k=i are then task e _iDistribute to the R of robot _j, make assigned _i=true, load ₁=true; If k is not equal to i, load ₁=false reads<e _k, R _j, n, level _k>, obtain level _k, if level _k=! Level _i, then e _iDistribute to the R of robot _j, make assigned _i=true, load ₁=true; Otherwise, make assigned _i=0; If load ₁=true makes n=n+1 turn to step 3);

6) judge level _I+1=level _i, if, then make p=i, i=i+1 turns to step 2), if not, then read<e _I1, R _j, n, level _I-1, 0>If, be empty, then make i=p+1, turn to step 2), otherwise, then make i=i1, execution in step 3) 4) 5)

7) termination routine.