CN202216696U - Coal mine disaster relief robot navigation device based on information integration - Google Patents

Abstract

The utility model discloses a coal mine disaster relief robot navigation device based on information integration. An information collection module collects information such as obstacle identifying information 6, road sign identifying information 7, scene identifying information 8, robot position speed information 9 and robot gesture information 10 which are under a coal mine, and the collected information is sent to an information processing module. The obstacle identifying information 6, the road sign identifying information 7 and the robot position speed information 9 are processed by an integration module 11 and a positioning module 12 which are in the information processing module and then achieve local path planning 13 together with the robot gesture information 10. The scene identifying information 8 is processed by an environment modeling module 14 in the information processing module to achieve overall path planning 15. Information of the local path planning 13 and information of the overall path planning 15 control a robot to perform autonomous navigation through a robot control module 16.

Description

Colliery disaster relief robot navigation device based on information fusion

Technical field

The utility model relates to a kind of robot navigation device, specifically is applied to colliery disaster relief robot, adopts the guider of multi-sensor information fusion technology.

Background technology

China is coal production and consumption big country, and in China's energy industry, coal accounts for about 70% in China's primary energy production and consumption structure.According to national working energy meeting prediction, to " 12 " end, the coal in China total quantity consumed will be above 4,000,000,000 tons, and wherein output in domestic will be above 3,800,000,000 tons, and coal in China output will account for 50% of global total production.Continuing to increase of coal production adds that the colliery natural conditions are poor, and technology is the reason of aspect such as backwardness and management shortcoming relatively, causes the mine disaster accident to take place frequently, and casualties is heavy.After explosion accident such as coal-mine gas, coal dust takes place, if there is kindling point, accidents such as subsequent explosion take place very easily, the personnel of speedily carrying out rescue work are difficult in the very first time and get into the disaster area, have incured loss through delay carrying out of rescue work.If robot is advanced into the down-hole earlier by the colliery disaster relief; The detection accident is destroyed and ambient conditions; Information feedback such as situation that gas density, harmful gas content, situation, scene stranded or wrecked personnel are caved in are to control center; Can improve the emergency repair ability like this, and aboveground decision-making is offered suggestions.

At present, navigate mode commonly used have GPS, simultaneously location and map structuring (simultaneous localization and mapping, SLAM), landmark navigation, inertial navigation, dead reckoning, infrared rays navigation, ultrasound wave navigation, vision guided navigation etc.

(1) GPS navigation: GPS navigation is a kind of overall navigational system that is applicable to outdoor robot, because colliery disaster relief robot is operated in the subsurface environment more, disaster relief robot can't adopt GPS to navigate under the coal mine.

(2) location and map structuring simultaneously: be in circumstances not known, to rely on the information that sensor obtained to carry out environmental modeling; Utilize the environmental map of being created to estimate pose, but the motion problems of this method control robot how also is not well solved.

(3) landmark navigation: utilize the road sign that is provided with in advance to navigate, because in a single day after the accident of blasting under the coal mine, the road sign that is provided with in advance is damaged most probably, this method is used under coal mine and is restricted.

(4) inertial navigation: be a kind of self-aid navigation that any external environment is carried out that do not rely on, obtain positional information, therefore have a cumulative errors through accumulation.

(5) dead reckoning: utilize magnetic compass can measure mobile robot's absolute direction, but bigger the mobile robot near the ferromagnetic material time error, also there is the low problem of measuring accuracy and response speed simultaneously.

(6) infrared rays navigation, ultrasound wave navigation: all utilize transponder pulse to navigate, because the absorptance of coal seam paired pulses is bigger, the loss that transmits is big, and emissive power is vulnerable to influence.

(7) vision guided navigation: have the information completely of obtaining, advantage such as investigative range is wide is occupied critical role in the robot navigation, and shortcoming is that real-time is relatively poor because the visual pattern processing time is long.

Under coal mine; Because the tunnel circumstance complication, the navigation signal that single-sensor collects has locality and one-sidedness, is difficult to the accuracy and the reliability of guarantee information; Causing robot to make accurately environmental information judges; And the airmanship that adopts multi-sensor information fusion can provide redundant information, the complementary information of disaster-stricken environmental characteristic under the coal mine simultaneously, but multiple information express-analysis current scene, thus guarantee that disaster relief robot accomplishes navigation task accurately and rapidly.In order to make full use of the information of various sensors, improve navigation accuracy, need to adopt blending algorithm that sensor information is carried out fusion treatment.Multisensor is the hardware foundation of information fusion, multi-sensor collection to information data be the process object that merges.Information fusion method commonly used has; Be applicable to method of weighted mean, EKF (the EKF:Extended Kalman Filter) method of dynamic environment, be applicable to that Bayes' assessment, the D-S of static environment proves rationalistic method, neural network and fuzzy reasoning method etc.

One Chinese patent application numbers 200910087237.7; Open day on November 11st, 2009; A kind of indoor movable robot real-time navigation method based on visual information correction is disclosed; This method adopts odometer and vision sensor Information Monitoring, and the structure Kalman filter is carried out data fusion, navigates.Under the coal mine after the calamity environmental uncertainty information more, rather dark, the quantity of information that odometer and vision sensor collect is limited, simultaneously there is the unmatched problem of information fusion in Kalman filter.

One Chinese patent application numbers 200810143134.3; Open day on January 28th, 2009; Disclose under a kind of circumstances not known mobile robot's multirow for merging automatic navigation method, this method is obtained current position angle in real time according to target and mobile robot's relative position, obtains distance parameter according to the barrier situation; Utilize fuzzy controller and environment identification controller to analyze data, navigate.This method self-adaptation is strong, and the navigation reliability is high, because in the destructuring environment under the coal mine, uncertain information category is many, data volume is big, the guider of fuzzy control is applied in the disaster relief robot of colliery, and constringency performance is relatively poor.

The CUMT-II colliery disaster relief robot of China Mining University research adopts binocular vision and array ultrasonic sensor to carry out path planning, in conjunction with infrared sensor to the barrier emergency brake that happens suddenly.The unfailing performance of this navigate mode satisfies the requirement under the coal mine.But how the inconsistency to the information of a plurality of sensors merges, and to reduce observational error be the key issue that disaster relief robot navigation device in colliery need solve with merging error.

It is thus clear that existing airmanship is used in the destructuring environment under coal mine and also need be improved, therefore, the guider that research is fit to disaster relief robot under the coal mine has great importance.

Summary of the invention

In order to satisfy the demand of destructuring environment under the coal mine, the utility model provides a kind of colliery disaster relief robot navigation device based on information fusion.The utility model is compared with guider in the past; Adopted ultrasound wave navigation, vision guided navigation and inertial navigation to design the guider of the Multi-sensor Fusion that is applicable to colliery disaster relief robot; Improved neural network-EKF method (NNEKF:Neural network Extended Kalman Filter) has been proposed; This algorithm can solve the unmatched problem of information fusion effectively, has improved the convergence and the real-time of device, has reduced error.

Below the method for the utility model is discussed.

A kind of colliery disaster relief robot navigation device based on information fusion; It is characterized in that; The position and speed information of the barrier identifying information under the coal mine, landmark identification information, scenery identifying information, robot and the attitude information of robot carry out information acquisition by information acquisition module; The information of gathering is sent to message processing module, wherein, and the Fusion Module in the position and speed information via message processing module of barrier identifying information, landmark identification information, robot, the processing of locating module; Realize local paths planning with the attitude information of robot; The scenery identifying information is realized global path planning through the processing of the environmental modeling module in the message processing module, and local paths planning information and global path planning information are through the control robot control module, thereby the realization control robot is carried out independent navigation.

Described guider, said information acquisition module further comprise ultrasonic wave module, vision module and inertia module.

Described guider, said ultrasonic wave module further comprise three pairs of ultrasonic transducers of the dead ahead that is placed in robot respectively, left front, right front, gather barrier identifying information, landmark identification information.

Described guider, said vision module adopts the trinocular vision sensor, gathers barrier identifying information, landmark identification information, scenery identifying information.

Described guider, said inertia module further comprises accelerometer, optical fibre gyro, accelerometer collection position velocity information, optical fibre gyro is gathered attitude information.

Described guider, the navigation information of ultrasonic wave module and vision module collection is revised the navigation information of inertia module collection.

Described guider further comprises explosion-proof module, and explosion-proof module adopts the explosion-proof mode of mixing of malleation intrinsic safety type.

Compared with prior art, the advantage of the utility model is:

(1) the utility model has adopted the colliery disaster relief robot navigation device based on the Multi-sensor Fusion of ultrasound wave navigation, vision guided navigation and inertial navigation; Fully merged the advantage separately of ultrasound wave navigation, vision guided navigation and inertial navigation; Improve the reliability of guider, strengthened the confidence level of data and the resolving power of device.

(2) the utility model has adopted the explosion-proof design of mixing of malleation intrinsic safety type, and the light weight of mixing explosion-proof design had both guaranteed the safety of guider to make robot have good exercise performance again.

(3) the utility model has adopted improved NNEKF algorithm, has improved the matching of convergence and information fusion, has remedied the shortcoming of the real-time difference of EKF algorithm when data volume increases simultaneously.

(4) the utility model utilizes BP neural network A to carry out the characteristic layer coupling and merges in improved NNEKF algorithm, has improved the matching degree of data fusion.

(5) the utility model is in improved NNEKF algorithm; The evaluated error of revising as input value; Return among the BP neural network A; Revise, can improve convergence of algorithm speed.

Description of drawings

Fig. 1 is a colliery disaster relief robot Multi-sensor Fusion guider structured flowchart

Fig. 2 is improved NNEKF algorithm composition frame chart

Fig. 3 is three layers of BP neural network structure figure

Fig. 4 is that improved NNEKF algorithm is realized block diagram

Fig. 5 is EKF algorithm simulating figure as a result

Fig. 6 is improved NNEKF algorithm simulating figure as a result

Among the figure, 1, ultrasonic wave module; 2, vision module; 3, inertia module; 4, accelerometer; 5, optical fibre gyro; 6, barrier identifying information; 7, landmark identification information; 8, scenery identifying information; 9, position and speed information; 10, attitude information; 11, merge; 12, location; 13, local paths planning; 14, environmental modeling; 15, global path planning; 16, robot control module; 17, input; 18, BP neural network A; 19, extended Kalman filter (EKF); 20, BP neural network B; 21, output; 22, input layer; 23, hidden layer; 24, output layer.

Embodiment

Following embodiment will further specify the utility model, and embodiment should not be regarded as the scope of restriction the utility model.Elaborate below in conjunction with accompanying drawing and embodiment working method to the utility model.

As shown in Figure 1, be colliery disaster relief robot Multi-sensor Fusion guider structured flowchart.

Ultrasonic wave module 1 is to utilize the mistiming between transmitted wave and the reception ripple to find range.

Measure equation is:

wherein; D is the distance of robot and tested barrier; C is the velocity of propagation of sound wave in medium, and T is transmitted wave and receives the mistiming between the ripple.

Adopt three pairs of ultrasonic transducers, be placed in dead ahead, left front, the right front of robot respectively.The advantage of ultrasound wave navigation is that the distance measuring method principle is simple, and finding range is wide, can be from several centimetres to tens meters, and information processing rate is fast, and real-time control is easy to do.Shortcoming is the wave beam broad, and its resolving power receives serious limit, and can only collect range information; Can not obtain information such as border, shape details; Because it is the dash current of emission is prone to cause gas subsequent explosion in the coal mine, therefore higher simultaneously to the explosion-proof designing requirement of sensor.

Vision module 2 adopts the trinocular vision navigation.The trinocular vision navigation can calculate the degree of depth of environment according to the parallax of the imaging of same scenery on three ccd video cameras, obtains steric information.Under coal mine among the disaster relief robot navigation, the vision sensor place that down-hole light is dark in the colliery descends the quality of image, can adopt method such as mine lamp illumination to compensate its defect.

Inertia module 3 relies on the airborne equipment of self to accomplish independent navigation fully, does not rely on the information of any outside, and good concealment is applicable under the coal mine bad working environment after the calamity.Inertial navigation carries out integration with it to the time through the linear acceleration and the angular acceleration of inertia device robot measurement at inertial coordinates system, can in time coordinate system, obtain speed, deviation angle and the positional information etc. of robot.Because inertial navigation obtains position and directional information through accumulation, the precision of measuring in the course of the work can descend, and cumulative errors constantly increases.Accelerometer 4 and optical fibre gyro 5 inertia devices are applicable to the coal mine operation environment.

Barrier identifying information 6 is gathered by ultrasonic wave module 1 and vision module 2 with landmark identification information 7; Scenery identifying information 8 is gathered by vision module 2; Position and speed information 9 is gathered by the accelerometer 4 of inertia module 3, and attitude information 10 is gathered by the optical fibre gyro 5 of inertia module 3.

Message processing module merges barrier identifying information 6, landmark identification information 7, position and speed information 9 processes 11, locatees 12; With the attitude information 10 common local paths plannings 13 of realizing of reflection robot the present situation, scenery identifying information 8 is realized global path planning 15 through environmental modeling 14.

Local paths planning 13 and the processing of global path planning 15 information through robot control module 16, control robot is carried out independent navigation.Ultrasonic wave module 1 is revised the navigation information of inertia module 3 collections with the navigation information that vision module 2 is gathered simultaneously.

Fully merged the advantage separately of ultrasound wave navigation, vision guided navigation and inertial navigation based on the multi-sensor fusion technology of ultrasound wave navigation, vision guided navigation and inertial navigation; Improve the reliability of guider, strengthened the confidence level of data and the resolving power of device.

The explosion-proof modular design of colliery disaster relief robot navigation device reaches as far as possible that volume is little, the characteristics of light weight, and the machine talent has good exercise performance like this.Take all factors into consideration above factor, adopt the explosion-proof design of malleation intrinsic safety type.Specific embodiments is following:

(1) the ultrasound wave receiving sensor of ultrasonic wave module 1, ultrasonic emitting sensor; The CCD camera of vision module 2; The accelerometer 4 of inertia module 3, optical fibre gyro 5, these environment detection equipment directly contact with external environment, are designed to essential explosion-proofing safe mode.

(2) other processing of circuit module, robot control module 16 are arranged in the pressurized enclosure, by compressor protection gas are provided in pressurized enclosure, keep air pressure inside greater than ambient air pressure, and damp gets in the pressurized enclosure in the prevention external environment.

As shown in Figure 2, be improved NNEKF algorithm composition frame chart.

Under destructuring environment under the coal mine; The uncertain characteristics that information category is many, data volume is big; The algorithm that adopts counterpropagation network (BP:back propagation network) to combine with the EKF method, and algorithm improved, improved NNEKF algorithm has been proposed.Information fusion mainly merges on data Layer, three different levels of characteristic layer and decision-making level.Because there is the big shortcoming of data processing amount in data Layer, does not carry out the fusion of data Layer, and adopt amalgamation mode with characteristic layer and decision-making level's combination.

At first utilize the huge advantage of BP neural network in study, classification and optimization, can the nonlinear transformations of uncertain environment under the coal mine through BP neural network A18 study reasoning, be carried out the characteristic layer coupling and merge; The algorithm that utilizes extended Kalman filter (EKF) 19 and BP neural network A18, BP neural network B20 to combine again carries out decision-making level and merges, and obtains exporting the result.Quoting of BP neural network not only can be improved the matching of convergence and information fusion, remedied the shortcoming of the real-time difference of EKF algorithm when data volume increases simultaneously.

As shown in Figure 3, be three layers of BP neural network structure figure.

The BP neural network is the feed-forward type neural network, comes the constantly weights of adjustment network through method of steepest descent, makes error sum of squares minimum.At present, most typical in the application is three layers of BP neural network structure, comprises input layer 22, hidden layer 23, output layer 24.

If the device input vector is X=[x ₁, x ₂..., x _n] ^T, be T=[t through the output vector of hidden layer 23 ₁, t ₂..., t _m] ^T, the device output vector is Y=[y ₁, y ₂..., y _l] ^T, desired output is D=(d ₁, d ₂..., d _l) ^T, the weight matrix between input layer 22 and the hidden layer 23 is a=[a ₁, a ₂..., a _m] ^T, the weight matrix between hidden layer 23 and the output layer 24 is b=[b ₁, b ₂..., b _l] ^T

In the dynamic navigation process of colliery disaster relief robot, along with the motion of CCD, the movement velocity of unique point in the CCD coordinate system is expressed as

Wherein: the unique point that x ', y ', z ' represent respectively to be recorded by ultrasound wave is along the linear velocity of coordinate axis X, Y, Z-direction; θ ' _f, φ ' _fRepresent the angular velocity of CCD respectively, measure the angular velocity of robot by gyroscope around X, Y, Z axle

θ ', φ ' obtain through conversion; X ' _f, y ' _f, z ' _fRepresent the linear velocity of CCD respectively along X, Y, Z-direction; (x, y, z) position of representation feature point in coordinate system.

For the colliery disaster relief robot with ultrasound wave navigation, vision guided navigation and inertial navigation unit, definition device state X is 16 dimensional vectors, and output state Y is 13 dimensional vectors that the measured value of guider constitutes.

Wherein: " representation feature point is along the acceleration of coordinate axis X, Y, Z-direction, q respectively for x ", y ", z ₁, q ₂, q ₃, q ₄The hypercomplex number of mobile robot's direction is described in expression respectively.

The purpose of BP neural network is exactly to ask the minimum value of error energy,

$E=\mathrm{\Σ}\frac{1}{2}{(Y-D)}^2---\left(3\right)$

Even (3) value of formula is minimum.The learning process of BP neural network comprises forward-propagating and error back propagation two parts.Forward-propagating is meant the processing of input signal through input layer 22, hidden layer 23, to output layer 24 outputs.Output signal Y and desired output D are not inconsistent, and then change backpropagation over to, and promptly output error signal arrives input layer 22 through hidden layer 23.Backpropagation can be assigned to error signal in each unit of each layer, is equivalent to the error signal of each unit of each layer is revised, and comes down to the process of a weights correction, thereby makes error energy E reach minimum.

As shown in Figure 4, be that improved NNEKF algorithm is realized block diagram.

Step 1: utilize input sample database training BP neural network A, utilize the gain that is input as predicted state and current state extended Kalman filter (EKF) of learning sample, be output as expected data training BP neural network B;

Step 2: the original signal that guider is collected is as input signal, and A learns reasoning by the BP neural network, and the output signal of BP neural network A is as the input signal of extended Kalman filter (EKF);

Step 3: the state equation of device and output equation are nonlinear properties, launch to be transformed into linear signal through Taylor's formula;

After the signal of input was handled through BP neural network A, the state equation X (k+1) and the output equation Y (k) that establish device did

$\left\{\begin{array}{c}X(k+1)=A[x\left(k\right),k]+B[x\left(k\right),k]W\left(k\right)\\ T\left(k\right)=C[x\left(k\right),k]+V\left(k\right)\end{array}\right.---\left(4\right)$

Wherein: A is that 13 dimensions can little vector equation, the state matrix of indication device; B is that 13 dimensions can little vector equation, the mapping matrix of expression noise auto levelizer; C is that 13 dimensions can little vector equation, the indication device output matrix; W and V are that relatively independent, average is zero white Gaussian noise vector, respectively the tracing device noise with measure noise.

A and C are non-linear matrix.The EKF algorithm is exactly to be launched into Taylor series to non-linear matrix A and C around filter value

and predicted value

respectively; Only keep the item below the secondary, obtain inearized model.

Carry out Taylor expansion to the A in the state equation around filter value

,

$X(k+1)\≈A[\hat{X}\left(k\right|k),k]+\frac{\∂A}{\∂X}{|}_{X\left(k\right)=\hat{X}\left(k\right|k)}[X\left(k\right)-\hat{X}\left(k\right|k)]+B\left[\hat{X}\left(k\right|k)\right]W\left(k\right)---\left(5\right)$

Order $\mathrm{\Γ}[k+1|k]=\frac{\∂A}{\∂X}{|}_{X\left(k\right)=\hat{X}\left(k\right|k)},$ $g\left(x\right)=A[\hat{X}\left(k\right|k),k]-\frac{\∂A}{\∂X}{|}_{X\left(k\right)=\hat{X}\left(k\right|k)}\hat{X}\left(k\right|k),$ Then

$X(k+1)=\mathrm{\Γ}[k+1|k]X\left(k\right)+g\left(x\right)+B[\hat{X}\left(k\right|k)]W\left(k\right)---\left(6\right)$

Carry out Taylor expansion to the C in the output equation around predicted value ,

$Y\left(k\right)\≈C[\hat{X}\left(k\right|k-1),k]+\frac{\∂C}{\∂X}{|}_{X\left(k\right)=\hat{X}\left(k\right|k-1)}[X\left(k\right)-\hat{X}\left(k\right|k-1)]+V\left(k\right)---\left(7\right)$

Order $P\left(k\right)=\frac{\∂C}{\∂X}{|}_{X\left(k\right)=\hat{X}\left(k\right|k-1)},$ $h(k+1)=C[\hat{X}\left(k\right|k-1),k]-\frac{\∂C}{\∂X}{|}_{X\left(k\right)=\hat{X}\left(k\right|k-1)}\hat{X}\left(k\right|k-1),$ Then

Y(k)＝P(k)X(k)+h(k)+V(k) (8)

Formula (6) and formula (8) have constituted the state equation of device and the inearized model of output equation.

Step 4: time renewal and test through extended Kalman filter (EKF) are upgraded, and get the Linear Estimation of the state equation X (k+1) of auto levelizer

Optimum Kalman filtering problem is known output sequence Y (0); Y (1); Y (k+1); The Linear Estimation that X (k+1) found out in requirement makes the variance of evaluated error

minimum, promptly

$E\left[\tilde{X}(k+1|k+1){\tilde{X}}^T(k+1|k+1)\right]=\min ---\left(9\right)$

$\hat{X}(k+1|k+1)=\hat{X}(k+1|k)+K(k+1)\{Y(k+1)-C[\hat{X}(k+1|k),k+1\left]\right\}---\left(10\right)$

Wherein: K (k+1) is an optimum gain battle array undetermined.

$\hat{X}(k+1|k)=A[\hat{X}\left(k\right|k),k]---\left(11\right)$

K (k+1)=S (k+1|k) C ^T(k+1) [C (k+1) S (k+1|k) C ^T(k+1)+R _K+1] ^-1(R is the positive definite matrix of symmetry)

(12)

$S(k+1|k)$

$=E\left[\tilde{X}(k+1|k){\tilde{X}}^T(k+1|k)\right]$

$=E\left\{\right[\mathrm{\Γ}(k+1,k)\tilde{X}\left(k\right|k)+B(k+1,k)W\left(k\right)\left]\right[\mathrm{\Γ}(k+1,k)\tilde{X}\left(k\right|k)+B(k+1,k)W\left(k\right){]}^T\}$ (13)

$=\mathrm{\Γ}(k+1,k)S\left(k\right|k){\mathrm{\Γ}}^T(k+1,k)+B(k+1,k)Q_k{B}^T(k+1,k)$

(Q is the non-negative definite matrix of symmetry)

Estimation error variance does

$S(k+1|k+1)=E\left[\tilde{X}(k+1|k+1){\tilde{X}}^T(k+1|k+1)\right]---\left(14\right)$

$=[I-K(k+1)C(k+1)S(k+1|k)]$

Wherein: the filter value initial value is

Step 5: after Linear Estimation

was handled through BP neural network B, the Linear Estimation that obtains revising

was as the output signal of device;

The evaluated error of revising

${\tilde{X}}^{\′}(k+1|k+1)=\hat{X}(k+1|k+1)-{\hat{X}}^{\′}(k+1|k+1)---\left(15\right)$

Step 6: be input to BP neural network A to the error signal of revising

, it is carried out retraining.

Like Fig. 5, shown in Figure 6, be respectively EKF algorithm simulating figure, improved NNEKF algorithm simulating figure as a result as a result.

The colliery disaster relief robot navigation device that combines with ultrasound wave navigation, vision guided navigation and inertial navigation is an example, respectively EKF algorithm and improved NNEKF algorithm is compared emulation experiment.To improved NNEKF algorithm simulating the time; Consider classification and the counting yield and the accuracy of colliery disaster relief robot navigation device identifying object; The BP neural network structure of in experiment, selecting for use adopts 5 * 8 * 2, i.e. 5 neurons of input layer, 8 neurons of hidden layer, 2 neurons of output layer.Suppose that the desired output signal is a segment length and is the straight line path of 10m, initial velocity V=0.5m/s establishes time t on interval [0,19.9], and SF is 5HZ, sample points N=100, and promptly BP neural network and EKF are through 100 times iterative learning.Can find out through simulation result:

(1) after the information via EKF algorithm output that navigation sensor is gathered, the error of output trajectory is between-0.2 to 0.2, and the convergence of output trajectory is relatively poor.

(2) after the improved NNEKF algorithm output of information via that navigation sensor is gathered, the error of output trajectory is stable fluctuation between-0.03 to 0.03, and visible improved NNEKF algorithm is higher than EKF algorithm navigation accuracy, better astringency; After BP neural network A carries out the fusion of characteristic layer coupling, improved the matching degree of data message; The error ratio of preceding 5 nodes is bigger simultaneously, because the BP neural network has the process of individual learning training, behind t=0.2s * 5=1s; Algorithm just is in steady state (SS), and visible improved NNEKF convergence of algorithm speed is very fast, after the 1s; The output signal just can be followed the tracks of target, and real-time is better.

(3) under coal mine, in the practical application, should combine the concrete condition of tunnel environment, speed is set, generally get between the 0.1m/s to 0.6m/s.Excessive velocities can also can increase the error of navigation, thereby influences the accuracy of guider.

Claims (7)

1. colliery disaster relief robot navigation device based on information fusion; It is characterized in that; The position and speed information of the barrier identifying information under the coal mine, landmark identification information, scenery identifying information, robot and the attitude information of robot carry out information acquisition by information acquisition module; The information of gathering is sent to message processing module, wherein, and the Fusion Module in the position and speed information via message processing module of barrier identifying information, landmark identification information, robot, the processing of locating module; Realize local paths planning with the attitude information of robot; The scenery identifying information is realized global path planning through the processing of the environmental modeling module in the message processing module, and local paths planning information and global path planning information are through the control robot control module, thereby the realization control robot is carried out independent navigation.

2. guider according to claim 1 is characterized in that, said information acquisition module further comprises ultrasonic wave module, vision module and inertia module.

3. guider according to claim 2 is characterized in that, said ultrasonic wave module further comprises three pairs of ultrasonic transducers of the dead ahead that is placed in robot respectively, left front, right front, gathers barrier identifying information, landmark identification information.

4. guider according to claim 2 is characterized in that, said vision module adopts the trinocular vision sensor, gathers barrier identifying information, landmark identification information, scenery identifying information.

5. guider according to claim 2 is characterized in that, said inertia module further comprises accelerometer, optical fibre gyro, accelerometer collection position velocity information, and optical fibre gyro is gathered attitude information.

6. guider according to claim 2 is characterized in that, the navigation information of ultrasonic wave module and vision module collection is revised the navigation information of inertia module collection.

7. guider according to claim 1 is characterized in that, further comprises explosion-proof module, and explosion-proof module adopts the explosion-proof mode of mixing of malleation intrinsic safety type.

Images