CN101354587A - Mobile robot multi-behavior syncretizing automatic navigation method under unknown environment - Google Patents

Abstract

The invention discloses a multi-behavior combining auto-navigation method for a mobile robot in an unknown environment. The method is characterized in that the method comprises the following steps that (1) current azimuth angle is obtained in real time according to the relative positions of an object and the mobile robot, and a plurality of distance parameters are obtained in real time according to the status of obstacles around the mobile robot; (2) a multi-output support vector machine fuzzy controller outputs a corner value Theta i and a velocity value vi according to the obtained azimuth angle and distance parameters, wherein i is equal to 1, 2 or 3; (3) a multi-output support vector machine environmental-identification controller inputs signals and outputs weight parameters wi of three subbehaviors according to input signals of the azimuth angle and the distance parameters, wherein i is equal to 1, 2 or 3; and (4) current corner value Theta i and velocity value vi of the mobile robot used for navigation are calculated according to the formula. The multi-behavior combining auto-navigation method adopts intelligent control strategy, and has the advantages of strong self adaptation, high navigation reliability and excellent effect.

Description

Mobile robot's multirow is for merging automatic navigation method under a kind of circumstances not known

Technical field

The invention belongs to robot navigation and field of intelligent control, relate under a kind of circumstances not known mobile robot's multirow for merging automatic navigation method.

Background technology

Over nearly 20 years, artificial intelligence technology and fast development of computer technology, autonomous intelligence mobile robot's research has obtained great concern.Intelligent mobile robot is widely used in industries such as industrial or agricultural, communications and transportation, military affairs, health care at present, to solve work problem and the human hard work of replacement under the hazardous environment.For improving ability to work and the range of application of mobile robot under circumstances not known, the research of mobile robot's independent navigation is a crucial difficult problem of being badly in need of solution.

For the mobile robot, homing capability is one of its most important function, robot at first require to avoid dangerous situation as the collision etc., robot is stayed under the safe operating environment; Secondly need possess the ability that is accomplished to a certain ad-hoc location execution particular task in the environment.Common navigation control method mainly is divided into two big classes at present: behavior control and potential field air navigation aid.The behavior Navigation Control is supposed known machine people's reference position and expectation target position usually, robot makes a strategic decision according to the sensor information of obtaining in the current subrange (ranging information, visual informations etc. such as infrared, sonar, laser), changes steering angle and movement velocity and bumps with barrier in expectation target direction running process avoiding.Such heuristic approach is simply effective, is used widely.

At present, for realizing the highly machine people control under the complex environment, fuzzy control, neural network, genetic algorithm scheduling theory are introduced between the design of sub-behavior controller and the sub-behavior and coordinate, the research of convergence strategy, but still lack reliability height, solution that adaptivity is strong.

Summary of the invention

Technical matters to be solved by this invention is to overcome the deficiencies in the prior art, provide under a kind of circumstances not known mobile robot's multirow for merging automatic navigation method, the mobile robot utilizes sonar, electronic compass sensor to obtain environmental information, finish environmental modeling, understand environment, thereby improve mobile robot's independent navigation ability.

Technical solution of the present invention is as follows:

Mobile robot's multirow is characterized in that for merging automatic navigation method under a kind of circumstances not known, may further comprise the steps:

1) relative position according to target and mobile robot obtains current position angle in real time, obtains a plurality of distance parameters in real time according to mobile robot's peripheral obstacle situation;

2) based on the many output support vector machine fuzzy controller that marches on towards target, Yan Qiang walks and keep away three sub-behaviors of barrier according to the position angle and a plurality of distance parameter output corner value θ that obtain _iWith velocity amplitude v _i, i=1,2,3;

3) based on march on towards target, Yan Qiang walks and keep away the barrier three sub-behaviors many output support vector machine environment identification controller according to position angle and a plurality of distance parameter input signal, output three sub-behaviors weight parameter w _i, i=1,2,3;

4) calculate current corner value θ that is used to navigate of mobile robot and velocity amplitude v according to following formula:

$\mathrm{\θ}=\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{w}_i\×{\mathrm{\θ}}_i,$ $v=\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{w}_i\×v_i.$

Described step 2) in, the employing electronic compass obtains the position angle ω between mobile robot and the impact point, adopts the ultrasonic ranging sensor of 8 ring-types configurations to obtain 8 distance parameter d1～d8; Wherein d1, d2, d3} are the left-hand distance, d4, d5} are the forward direction distance, d6, d7, d8} are the dextrad distance;

The obstacle signal left-hand obstacle Left_obs of described many output support vector machine fuzzy controllers, forward direction obstacle Front_obs and dextrad obstacle Right_obs are defined as:

Left_obs＝Min{d1，d2，d3}；Front_obs＝Min{d4，d5}；Right_obs＝Min{d6，d7，d8}。

For described many output support vector machine fuzzy controllers, be input as left-hand obstacle Left_obs, forward direction obstacle Front_obs and dextrad obstacle Right_obs along what the hedge behavior was walked and kept away to wall, what march on towards the sub-behavior of target is input as position angle ω; Controller is output as two, corner value θ _iWith velocity amplitude v _i, i=1,2,3, i be 1,2,3 correspondence marches on towards target, Yan Qiang walks respectively and keep away the barrier three sub-behaviors; Left-hand obstacle Left_obs, forward direction obstacle Front_obs, dextrad obstacle Right_obs adopt { far away, near, very near } promptly { fuzzy set that three fuzzy language variablees of Very near} are formed represents that position angle ω adopts { big a left side for Far, Near, a middle left side, a left side, forward, the right side, the middle right side, big right } i.e. { far-left, medium-left, left, head-on, right, medium-right, seven fuzzy language variablees of far-right} are represented; The output fuzzy set rotational angle theta of behavior fuzzy controller _iWith fuzzy set negative big, and negative in, negative little, zero, just little, the center, honest } promptly NB, NM, NS, ZE, PS, PM, PB} represent, speed v _iPromptly { Slow} represents for Fast, Medium with fuzzy set { fast, in, slow }.

For described many output support vector machine fuzzy controllers, utilize many output support vector machine to return, the rule that fuzzy expert knowledge is definite is as sample, in the permissible error scope, carry out approximation of function, select the sample that becomes support vector, thereby weed out the redundancy rule in the rule base, obtain fitting function.

Described many output support vector machine environment identification controllers be input as distance parameter { d1, d2, d3, d4, d5, d6, d7, d8} and position angle ω; Output layer be current environment respectively with the matching degree f of three sub-behaviors _n(n=1,2,3), the matching degree sum equals 1, at first obtain environmental information during training as input, and be the desired outputs of each sub-behavior value with matching degree current environment that distribute as many output support vector machine environment identification controllers according to expertise, obtain a series of training sample thus and these many output support vector machine environment identification controllers are trained; Training is finished resulting network structure and is described many output support vector machine environment identification controllers; The weight parameter w of three sub-behaviors then _i, i=1,2,3 are:

[w ₁，…，w ₃] ^T＝[f ₁，…，f ₃] ^T， $\underset{n=1}{\overset{3}{\mathrm{\Σ}}}{w}_n=1,$ w _n∈[0，1]。

Beneficial effect:

Compared with prior art, advantage of the present invention just is:

1, adopts many output support vector machine fuzzy controllers, can reduce redundancy rule effectively, omit the reverse gelatinization process, simplified calculating, optimized fuzzy controller.

2, utilize many output support vector machine identification environment, dynamically determine the matching degree of sub-behavior and current environment of living in according to actual environment, thereby determine sub-behavior output shared weight in whole navigation behavior output, improved the confidence level of navigation behavior output.The convergence strategy that the proposition multirow is has further been optimized the navigation weighting parameter.

3, modular construction is conveniently added other sub-behavior, has improved the adaptivity of system.

Experimental result (seeing embodiment) has verified that also the present invention can realize mobile robot's navigation reliably.

Description of drawings

Fig. 1 is that mobile robot's multirow is fusion air navigation aid overview flow chart under the circumstances not known of the present invention;

Fig. 2 is mobile robot's forward direction sonar ring (ultrasonic ranging sensor) configuration schematic diagram;

Fig. 3 is sub-behavior structure of fuzzy controller synoptic diagram;

Fig. 4 is the input and output fuzzy membership function;

Fig. 5 exports the support vector machine program flow diagram for off-line training more;

Fig. 6 is current environment and sub-behavior matching degree training sample;

Fig. 7 is figure synoptic diagram ideally;

Fig. 8 is that the Navigation Control to target (goal) is tested comparing result ((a) A in the environment 1 ^*The path planning of algorithm (b) this method is to pursuit path (c) this method independent navigation path of path planning);

Fig. 9 is that the Navigation Control to target (goal) is tested comparing result ((a) A in the environment 1 ^*The path planning of algorithm (b) this method is to pursuit path (c) this method independent navigation path of path planning);

Figure 10 is that the Navigation Control to target (goal) is tested comparing result ((a) A in the environment 2 ^*The path planning of algorithm (b) this method is to pursuit path (c) this method independent navigation path of path planning);

Embodiment

With embodiment enforcement of the present invention is described with reference to the accompanying drawings.

Below with reference to accompanying drawing and specific implementation process the present invention is described in further details:

Embodiment 1:

System construction drawing of the present invention is (FSVM is a fuzzy support vector machine among Fig. 1) as shown in Figure 1, input signal is position angle and 8 distance parameters, the navigation behavior is divided into marches on towards target, Yan Qiang and walk, keep away three sub-behaviors of barrier, sensor obtains information and sends into sub-behavior controller and environment identification controller respectively.The output of sub-behavior controller is the rotational angle theta and the movement velocity v of robot, and environment identification controller is output as the matching degree of sub-behavior and current environment.It is the output of whole navigational system that sub-behavior controller output is exported the sum of products with corresponding environment identification controller, and the control mobile robot turns to and speed.Rotational angle theta and movement velocity v are used for realizing the navigation behavior, and wherein rotational angle theta is used to change the direction of motion of robot.The mobile robot adopts ultrasonic ranging sensor, electronic compass to finish environmental modeling and the sub-behavior of navigating, and its front end is equipped with eight sonar ranging sensors.The position of mobile robot's sonar sensor is among Fig. 2: ± 10 °, ± 30 °, ± 50 ° and ± 90 °, be used to survey the information of barrier on the direction separately.The ranging information of sonar is divided into a left side, preceding, right three sectors, is designated as Left_obs respectively, Front_obs, and Right_obs is used to survey the obstacle information of these three directions.

1, exports support vector loom behavior fuzzy controller more

At first at sub-behavior design design fuzzy controller, many then output support vector machine return principle, adopt many output support vector machine to extract control law.

1) sub-behavior

Illustrate: the navigation behavior is divided into and marches on towards target, Yan Qiang walks, keeps away three sub-behaviors of barrier.Wherein march on towards goal behavior and be meant not consider whether the place ahead can pass through, a head for target place direction is advanced; Walk behavior along wall and be meant along corridor, wall etc. have longer linear edge to advance, and ignore target place direction; Keep away the barrier behavior and be meant that the barrier that gets around the place ahead static state continues to advance forward, and ignore target place direction.

2) sub-behavior fuzzy controller

Illustrate: utilize fuzzy theory for designing sub-behavior fuzzy controller.

Algorithm:

Usually electronic compass obtains position angle ω between robot and the impact point; Mobile robot's 8 sonar to measure barriers and the distance between the robot, and d1, d2, d3, d4, d5, d6, d7, d8} is divided into Left_obs according to formula (1), Front_obs, three groups of Right_obs, wherein Min () expression is got minimum value from a plurality of worthwhile.

Left_obs＝Min{d1，d2，d3}；Front_obs＝Min{d4，d5}；Right_obs＝Min{d6，d7，d8}(1)

The fuzzy controller that the present invention adopts is determined input according to the function of sub-behavior, is Left_obs,, the collocation of Front_obs, Right_obs and ω.Its input of sub-behavior that generally needs to determine range information must comprise Left_obs,, Front_obs, Right_obs, its input that needs to determine orientation between robot and the impact point must comprise ω, promptly walk, keep away the Left_obs that is input as of barrier behavior along wall,, Front_obs, Right_obs, march on towards the ω that is input as of goal behavior.Controller is output as two, i.e. rotational angle theta and speed v.Its structure is imported among the figure and is adopted dotted line as shown in Figure 3, and expression is selected input according to the needs of sub-behavioral function.Three groups of range informations are all available, and { fuzzy set that three fuzzy language variablees of Very near} are formed represents that azimuth information ω can use { far-left for Far, Near, medium-left, left, head-on, right, medium-right, seven fuzzy language variablees of far-right} are represented; { NB, NM, NS, ZE, PS, PM, PB} represent that { Slow} represents speed v for Fast, Medium with fuzzy set to the output fuzzy set rotational angle theta of behavior fuzzy controller with fuzzy set.The fuzzy membership function of input and output as shown in Figure 4.

Then at each behavior design fuzzy control rule, march on towards fuzzy control rule that goal behavior, Yan Qiang walk behavior and keep away the barrier behavior respectively shown in table 1, table 2, table 3.

Table 1 marches on towards the goal behavior fuzzy rule base

The measurement data array case	Rule numbers	ω	θ 1	v 1
					1	1	far-left	NB	Fast
2	2	medium-left	NM	Medium

3	3	left	NS	Medium
					4	4	head-on	ZE	Fast
5	5	right	PS	Medium
					6	6	medium-right	PM	Medium
7	7	far-right	PB	Fast

Table 2 is walked the behavior fuzzy rule base along wall

The measurement data array case	Rule numbers	Left_obs	Front_obs	Right_obs	θ 2	v 2
							1	1	Very near	Very near	Very near	NB	Slow
2	2	Very near	Very near	near	PM	Slow
							3	3	Very near	Very near	Far	PS	Slow
4	4	Very near	near	Very near	ZE	Slow
							5	5	Very near	near	near	ZE	Slow
6	6	Very near	near	Far	PS	Medium
							7	7	Very near	Far	Very near	ZE	Fast
8	8	Very near	Far	near	ZE	Fast
							9	9	Very near	Far	Far	ZE	Fast
10	10	near	Very near	Very near	NS	Slow
							11	11	near	Very near	near	NS	Slow
12	12	near	Very near	Far	PM	Medium
							13	13	near	near	Very near	ZE	Slow
14	14	near	near	near	ZE	Slow
							15	15	near	near	Far	PM	Medium
16	16	near	Far	Very near	ZE	Fast
							17	17	near	Far	near	ZE	Fast
18	18	near	Far	Far	ZE	Fast
							19	19	Far	Very near	Very near	NM	Medium
20	20	Far	Very near	near	NM	Medium

21	21	Far	Very near	Far	NM	Medium
							22	22	Far	near	Very near	ZE	Medium
23	23	Far	near	neaf	ZE	Medium
							24	24	Far	near	Far	ZE	Medium
25	25	Far	Far	Very near	ZE	Fast
							26	26	Far	Far	near	ZE	Fast
27	27	Far	Far	Far	ZE	Fast

Table 3 is kept away barrier behavior fuzzy rule base

The measurement data array case	Rule numbers	Left_obs	Front_obs	Right_obs	θ 3	v 3
							1	1	Very near	Very near	Very near	NB	Slow
2	2	Very near	Very near	near	PM	Slow
							3	3	Very near	Very near	Far	PM	Slow
4	4	Very near	near	Very near	NB	Slow
							5	5	Very near	near	near	PM	Medium
6	6	Very near	near	Far	PS	Fast
							7	7	Very near	Far	Very near	ZE	Medium
8	8	Very near	Far	near	ZE	Fast
							9	9	Very near	Far	Far	PS	Fast
10	10	near	Very near	Very near	NM	Slow
							11	11	near	Very near	near	NM	Slow
12	12	near	Very near	Far	PM	Slow
							13	13	near	near	Very near	NM	Medium
14	14	near	near	near	NM	Medium
							15	15	near	near	Far	PS	Medium
16	16	near	Far	Very near	NS	Medium
							17	17	near	Far	near	ZE	Fast
18	18	near	Far	Far	PS	Fast

19	19	Far	Very near	Very near	NM	Slow
							20	20	Far	Very near	near	NM	Slow
21	21	Far	Very near	Far	NM	Slow
							22	22	Far	near	Very near	NS	Medium
23	23	Far	near	near	NS	Medium
							24	24	Far	near	Far	NS	Medium
25	25	Far	Far	Very near	NS	Fast
							26	26	Far	Far	near	NS	Fast
27	27	Far	Far	Far	ZE	Fast

3) many output support vector machine extract control law

Illustrate: utilize many output support vector machine to return, the rule that fuzzy expert knowledge is definite is as sample, in certain permissible error scope, carry out approximation of function, select the sample that becomes support vector, thereby weed out the redundancy rule in the rule base, and the fitting function that obtains having better learning ability, promptly based on the fuzzy controller of many output support vector machine.Algorithm:

According to the input and output quantity of sub-behavior fuzzy controller, determine the input and output quantity of support vector machine.Be the input quantity that the input number of nodes amount of support vector machine equals sub-behavior fuzzy controller; The output node quantity of support vector machine equals the output quantity of sub-behavior fuzzy controller, is rotational angle theta _iAnd speed v _iSub-behavior fuzzy controller comprises a series of control laws of being made up of IF THEN, and a control law in the rule base has N bar rule corresponding to center of a sample's point in the rule base, will obtain N sample point.Then this N sample point adopted the support vector machine homing method, under appropriate accuracy ε, use this N of non-linear regression function match sample error freely, find the solution the quadratic programming optimal problem, find the support vector (s) in the sample, its pairing control law is useful rule, but not the pairing control law of the sample of support vector is the redundancy rule in the rule base, also obtains fitting function, i.e. support vector loom behavior fuzzy controller simultaneously.

Marching on towards the goal behavior fuzzy control rule with the support vector machine extraction is that example illustrates its process.

As shown in table 1 march on towards a control law in the goal behavior fuzzy rule base, 7 rules are arranged in this rule base, will obtain 7 sample points corresponding to center of a sample's point, as shown in table 4.

Table 4 marches on towards the goal behavior sample

The array of samples situation

Input

Output

	ω	(θ 1，v 1)
			1	far-left	(NB，Fast)
2	medium-left	(NM，Medium)
			3	left	(NS，Medium)
4	head-on	(ZE，Fast)
			5	right	(PS，Medium)
6	medium-right	(PS，Medium)
			7	far-right	(PB，Fast)

Then these 7 sample points are adopted the (input of support vector machine homing method, two outputs), (rule of thumb adjust in appropriate accuracy ε＞0, select ε=0.3 herein) down with these 7 samples of non-linear regression function match, the training program flow diagram as shown in Figure 5, map parameter c is the penalty factor and c＞0 of support vector machine.It is K (x that this method adopts the RBF kernel function _i, x)=exp (g|x-x _i| ²), x _iBe the support vector that will find the solution, the parameter g that reaches among Fig. 5 in the formula is the parameter and g＞0 of kernel function.

2, based on the environment identification of many output support vector machine

Illustrate: the Mobile Robotics Navigation task is decomposed into three sub-behaviors, how to determine the matching degree of mobile robot's environment of living in and sub-behavior, finally determines the weight of sub-behavior in navigational system output, is related to the performance of navigational system.The present invention adopts many output support vector machine to come the matching degree of identification environment and sub-behavior, improves the reliability of system.

Algorithm:

The input of many output support vector machine environment identification controllers has nine, promptly reflects sonar ranging information { d1, d2, d3, d4, d5, d6, d7, the angle ω between d8} and mobile robot and the target of mobile robot's peripheral obstacle information; Output layer is the matching degree f of environment and sub-behavior _n(n=1,2,3), the matching degree sum equals 1, and the quantity of output node is consistent with sub-behavior quantity.At first obtain environmental information during training as input, and according to expertise for the matching degree value of each sub-behavior distribution and the current environment of design as the desired outputs of exporting support vector machine environment identification controllers more, can obtain a series of training sample like this.The Mobile Robotics Navigation task is decomposed into and marches on towards goal behavior, keep away the barrier behavior and Yan Qiang walks behavior, the part training sample example of dividing gamete behavior and current environment matching degree as shown in Figure 6:

Among Fig. 6 (a), sonar sensor detects mobile robot's periphery clear, and the matching degree that marches on towards goal behavior and current environment is 1, and the matching degree of other sub-behavior and current environment is 0;

Among Fig. 6 (b), sonar sensor detects mobile robot the right barrier, and target is in its left front, and the matching degree that marches on towards goal behavior and current environment is 1, and the matching degree of other sub-behavior and current environment is 0;

Among Fig. 6 (c), sonar sensor detects mobile robot the right barrier, and target is in its right back, and the matching degree of walking behavior and current environment along wall is 1, and the matching degree of other sub-behavior and current environment is 0;

Among Fig. 6 (d), sonar sensor detects the mobile robot dead ahead barrier, and target is also in its dead ahead, and the matching degree that marches on towards goal behavior and current environment is 0.1, the matching degree of keeping away barrier behavior and current environment is 0.9, and the matching degree of walking behavior and current environment along wall is 0.

Fig. 6 abstract environment, can build as shown in Figure 7 actual environment and ecotopia as shown in Figure 8, robot places this known environment to obtain training sample.For Fig. 6 (a) target is placed different position ([90 ° of the position angle ω ∈ on every side of robot, 90 °], the sonar ranging scope is 0～3 meter), can obtain different measurement sample points according to the different distributions of barrier, impact point, wall, wherein a part of sample is as shown in table 5.

Table 5 environment and sub-behavior matching degree training part sample

(can be interpreted as a structure that is similar to neural network to support vector machine,, know input, output, send into the support vector machine training and get final product according to Fig. 5 as long as according to sample.Be before fuzzy rule as training sample, be that environmental information and weights are sent into the support vector machine training as sample now.) shown in support vector machine training process flow diagram support vector machine is trained, the network structure that obtains is many output support vector machine environment identification controllers.

3, behavior is merged

Because the fusion of the output of the navigational system sub-behaviors output that is all, the matching degree (f of environment and certain individual sub-behavior ₁, f ₂, f ₃) high more, the weights of corresponding sub-behavior output are just big more, shown in weights are calculated as follows:

[w ₁，…，w ₃] ^T＝[f ₁，…，f ₃] ^T， $\underset{n=1}{\overset{3}{\mathrm{\Σ}}}{w}_n=1,$ w _n∈[0，1](2)

The therefore last actual mobile robot's of outputing to of navigational system controlled quentity controlled variable rotational angle theta and movement velocity v are:

$\mathrm{\θ}=\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{w}_i\×{\mathrm{\θ}}_i,$ $v=\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{w}_i\×v_i---\left(3\right)$

θ wherein ₁, v ₁Represent that designed many output support vector machine marches on towards the sub-behavior fuzzy controller output of target, θ ₂, v ₂Represent the output that designed many output support vector is kept away hedge behavior fuzzy controller, θ ₃, v ₃Represent that designed many output support vector walks the output of sub-behavior fuzzy controller along wall.

Experimental result and analysis

For verifying the validity of the navigation control method that the present invention proposes, with the artificial object of Pioneer 2-DXE mobile apparatus, the Navigation Control experiment has been carried out in two following 3 target locations of varying environment, and wherein among Fig. 8 to Figure 10, the movement locus of robot is A among the figure (a) ^*The standard planning path of algorithm; Figure (b) is for being divided into the tracking test result (promptly environment is known at this moment) after the plurality of sub target to the standard planning path; Figure (c) does not rely on path planning for the unknown of supposition environment, and directly according under initial position and the target position information situation, the Navigation Control track of the inventive method is with the reliability and the adaptivity (A of verification method ^*Algorithm is a kind of method for searching path of classics, the a lot of nodes (providing node location information in program) that are provided with artificial between start node and destination node play the road sign effect, just a lot of individual sub-goals, sub-goal is many more, show that the information of telling robot is just many more, can think that like this environment just is cicada, can realize by the orientation and the coordinate of given sub-goal; And circumstances not known promptly only provides starting point and destination node, does not provide the position of intermediate node, therefore, for robot, is a circumstances not known).

From experimental result, this air navigation aid has very high precision and reliability to known paths, and three kinds of situations all can accurately be approached the standard planning path.Under circumstances not known, this air navigation aid has also shown extraordinary accuracy and to the adaptability of environment, all can finally reach the target location, in environment 2,, cause robot the rollback behavior to occur owing to lead to the appearance in back-shaped zone in the process of target location, path optimization's property is relatively poor, but this air navigation aid finally still can be broken away from local trap influence, finally reach the target location, so this air navigation aid has good adaptive faculty and reliability to varying environment.

Claims (2)

1. mobile robot's multirow is characterized in that for merging automatic navigation method under the circumstances not known, may further comprise the steps:

1) relative position according to target and mobile robot obtains current position angle in real time, obtains a plurality of distance parameters in real time according to mobile robot's peripheral obstacle situation;

2) based on the many output support vector machine fuzzy controller that marches on towards target, Yan Qiang walks and keep away three sub-behaviors of barrier according to the position angle and a plurality of distance parameter output corner value θ that obtain _iWith velocity amplitude v _i, i=1,2,3;

3) based on march on towards target, Yan Qiang walks and keep away the barrier three sub-behaviors many output support vector machine environment identification controller according to position angle and a plurality of distance parameter input signal, output three sub-behaviors weight parameter w _i, i=1,2,3;

4) calculate current corner value θ that is used to navigate of mobile robot and velocity amplitude v according to following formula:

$\mathrm{\θ}=\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{w}_i\×{\mathrm{\θ}}_i,$ $v=\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{w}_i\×v_i;$

Described step 2) in, the employing electronic compass obtains the position angle ω between mobile robot and the impact point, adopts the ultrasonic ranging sensor of 8 ring-types configurations to obtain 8 distance parameter d1～d8; Wherein d1, d2, d3} are the left-hand distance, d4, d5} are the forward direction distance, d6, d7, d8} are the dextrad distance;

The obstacle signal left-hand obstacle Left_obs of described many output support vector machine fuzzy controllers, forward direction obstacle Front_obs and dextrad obstacle Right_obs are defined as:

Left_obs＝Min{d1，d2，d3}；Front_obs＝Min{d4，d5}；Right_obs＝Min{d6，d7，d8}；

For described many output support vector machine fuzzy controllers, be input as left-hand obstacle Left_obs, forward direction obstacle Front_obs and dextrad obstacle Right_obs along what the hedge behavior was walked and kept away to wall, what march on towards the sub-behavior of target is input as position angle ω; Controller is output as two, corner value θ _iWith velocity amplitude v _i, i=1,2,3, i be 1,2,3 correspondence marches on towards target, Yan Qiang walks respectively and keep away the barrier three sub-behaviors; Left-hand obstacle Left_obs, forward direction obstacle Front_obs, dextrad obstacle Right_obs adopt { far away, near, very near } fuzzy set formed of three fuzzy language variablees represents that position angle ω adopts { a big left side, a middle left side, a left side, forward, the right side, the middle right side, big right } seven fuzzy language variablees represent; The output fuzzy set rotational angle theta of behavior fuzzy controller _iWith fuzzy set negative big, and negative in, negative little, zero, just little, the center, honest } expression, speed v _iWith fuzzy set { fast, in, slow } expression;

For described many output support vector machine fuzzy controllers, utilize many output support vector machine to return, the rule that fuzzy expert knowledge is definite is as sample, in the permissible error scope, carry out approximation of function, select the sample that becomes support vector, thereby weed out the redundancy rule in the rule base, obtain fitting function.

2. mobile robot's multirow is characterized in that for merging automatic navigation method under the circumstances not known according to claim 1, described many output support vector machine environment identification controllers be input as distance parameter { d1, d2, d3, d4, d5, d6, d7, d8} and position angle ω; Output layer be current environment respectively with the matching degree f of three sub-behaviors _n(n=1,2,3), the matching degree sum equals 1, at first obtain environmental information during training as input, and be the desired outputs of each sub-behavior value with matching degree current environment that distribute as many output support vector machine environment identification controllers according to expertise, obtain a series of training sample thus and these many output support vector machine environment identification controllers are trained; Training is finished resulting network structure and is described many output support vector machine environment identification controllers; The weight parameter w of three sub-behaviors then _i, i=1,2,3 are:

[w ₁，…，w ₃] ^T＝[f ₁，…，f ₃] ^T， $\underset{n=1}{\overset{3}{\mathrm{\Σ}}}{w}_n=1,$ w _n∈[0，1]。

Images