CN102401656A - Navigation algorithm of bionic robot for position cells - Google Patents

Abstract

The invention discloses a positional cell bionic robot navigation algorithm, A1, firstly, a wide-angle camera of a robot takes a panoramic picture, converts the panoramic picture into a gradient map, and then, a Gaussian difference filter is used for filtering the panoramic picture to detect characteristic points; carrying out logarithmic polar coordinate transformation on local areas near the characteristic points, wherein the characteristic points correspond to specific road signs; a2, acquiring an angle relative to the true north direction, i.e. azimuth, provided by a compass, for each landmark; the azimuth angle and the road marking jointly determine the position cell at the current moment; a3, determining a transfer matrix by the position cell at the current moment and the position cell at the previous moment; when the robot is exploring in an unknown environment, a transfer matrix is continuously formed, and the transfer matrix forms a cognitive map; a4, jointly determining a motion transformation matrix by the transfer matrix and the cognitive map, wherein the motion transformation matrix is responsible for sending out a motion command; the robot can be applied to intelligent ground cleaning robots, battlefield search and rescue robots and the like.

Description

A kind of position cell bio-robot navigation algorithm

Technical field

The present invention relates to a kind of position cell bio-robot navigation algorithm.Formulate autonomous operation robot navigation algorithm according to the navigation strategy of mammal hippocampus position cell.Can be used for the intelligent cleaning ground robot, battlefield search and rescue robot etc.

Background technology

Mammal (like rat, people etc.) cost great amount of time moves to another place from the three unities.Autotelic moving need be encoded to motivation, space.Animal is sane to the coding of space and motivation.About its Study on Mechanism artificial intelligence field there is important enlightenment.

" cognitive map " is meant the inherent expression way of animal to spatial information.Cognitive map can be expressed through a series of terrestrial reference.In the process of tens seconds to a few minutes, can be familiar with environment and accurately memory." position cell " is the cone neurone in the hippocampus, and the granting rate is higher when certain local location in the animal environment, and it is almost nil to leave this position granting rate.The position cell has following attribute:

1, in new environment, the position cell can be set up rapidly;

2, position wild (the position cell is provided corresponding position) is stable, the identical position of position cell coding likewise, in addition still stable behind the some months;

3, the open country, position does not tightly depend on visual information, and is effective at the environment of dark yet;

4, the displacement of road sign and rotation cause displacement and the rotation that the position is wild at a distance;

5, identical position cell possibly have diverse position wild in the different environment granting;

6, the position cell receives the modulation of direction cell to the end.

Summary of the invention

The present invention provides a kind of position cell bio-robot navigation algorithm according to the position cell coding strategy in the cerebral hippocampal.For reaching above purpose, the present invention takes following technical scheme to be achieved:

Landmark identification based on characteristic.At first utilize camera to obtain unique point, next utilizes unique point to obtain positional information.Add that through unique point positional information sets up road sign deflection network.Image is that the full shot with low resolution obtains.Image is removed brightness through gradient distribution and is disturbed.Gradient image and double gauss difference operator convolution are carried out feature identification.Road sign learning of neuron road sign.For each road sign, the relative angle of direct north is read by compass relatively.The vision system of this model provides " what " and " where " two kinds of information.

Two information fusion of road sign and deflection produce the position cell, and the reposition that in the process of exploring, occurs is with new neuron coding.In certain given position, a plurality of positions cellular activity, co.The density of terrestrial reference is relevant with robot residing environment position.Local fast in position angle change in location such as wall and doors, can acquire more terrestrial reference.After entire environment had all been learnt, environment can be covered by the position cell fully, and each position cell is corresponding with corresponding position.The position cell is the position that robot confirms oneself.

The transition matrix coding.For planned navigation task, must accomplish certain track.These tracks can be represented with the sequence of location point.Whole matrix (transition matrix) from present position to next position can be used for representing this track.Can from matrix, remove impossible track and partly reduce resource occupation.

After transition matrix is set up, form cognitive map.Cognitive map and transition matrix form the motion converter matrix jointly.The motion converter matrix is responsible for order is sent.For example, from the position A B has produced transitional cell AB to the position.This transitional cell and the direction from A to B interrelate.

The autonomous robot and the mankind seemingly also should be designed with motivation.Motivation can be certain will accomplishing of task, seeks power source charges when self electric quantity is not enough, gets back to the fixed location after perhaps finishing the work.

The invention discloses a kind of position cell bio-robot navigation algorithm.Formulate robot navigation's algorithm according to mammal based on the navigation strategy of hippocampus position cell.Can be applicable to is the intelligent cleaning ground robot, battlefield search and rescue robot etc.Can discern foreign environment automatically, have advantages such as the human intervention of need not, self-organization, self-adaptation.

Description of drawings

Fig. 1 hardware block diagram of the present invention;

Fig. 2 algorithm flow chart of the present invention;

The formation synoptic diagram of Fig. 3 road sign of the present invention-position angle cell.

Embodiment

Below in conjunction with specific embodiment, the present invention is elaborated.

Position of the present invention as shown in Figure 1 cell bio-robot navigation algorithm based on hardware, comprise: the wide-angle imaging head is used to obtain external image; Dsp chip is used to carry out the relevant algorithm of robot learning and cognitive environment of living in; Driving wheel, the motion command drive machines people who sends according to dsp chip moves.

Position cell biologically is only relevant with the position that biology is positioned, and all it doesn't matter with the speed of its travel direction and motion.The position cell of learning for mimic biology, in the software environment of DSP with variable analog position cell.

As shown in Figure 2, at first robot wide-angle imaging head picked-up distant view photograph in order to remove the interference of brightness, is translated into gradient map, uses the difference of gaussian wave filter to its filtering subsequently, the detected characteristics point.Near to the unique point regional area is done log-polar transform, can improve the recognition correct rate to small rotation (rotation) or yardstick (scale) variation.Because these unique points corresponding to specific road sign, therefore will be called road sign cell (landmark cells) corresponding to the variable of these unique points.

(be the position angle, azimuth), direct north is provided by compass to obtain angle with respect to direct north for each road sign.The visual field of 360 degree is by N _AzmIndividual position angle cell coding.The position cell of current time is confirmed on position angle and mark road jointly.

For planned navigation task, must accomplish certain track.These tracks can be represented with the sequence of location point.Whole matrix (transition matrix) from present position to next position can be used for representing this track.Can from matrix, remove impossible track and partly reduce resource occupation.The position cell of current time and the position cell in the previous moment are confirmed transition matrix jointly.

When robot in unknown the exploration time, constantly forms transition matrix in environment, transition matrix forms cognitive map; Transition matrix and cognitive map determine the motion converter matrix jointly; Output movement had been ordered a period of time, after transition matrix is set up, had just formed so-called cognitive map.Cognitive map can be thought the node that transfer vector and border are formed.Node self shifts weight and is made as 1, shifts weight to other node and is made as 0.9.Weight increases and reduces along with the path frequency of utilization.

The motion converter matrix is responsible for motion command is sent.For example, from the position A B has produced transitional cell AB to the position.This transitional cell and the direction from A to B interrelate.

Each motion command all is through certain direction position of reaching home from certain start position.For example, from the position A to the position B, produce order AB, also comprise the direction from A to B.

The autonomous robot and the mankind seemingly also should be designed with motivation.Motivation can be certain will accomplishing of task, seeks power source charges when self electric quantity is not enough, gets back to the fixed location after perhaps finishing the work.

Two information fusion in road sign and position angle produce road sign position angle fused cell, and this cell is the intermediate variable that produces the position cell, and its movable computing method are divided into three steps.At first obtain the road sign cell maximum activity

and all deflection cells maximum activity

secondly; Calculate the product of these two maximum activity; It is last to be defined as

, obtains the activity of road sign deflection fused cell:

X ^Prd(t+1)＝[X ^Prd(t)+p] ⁺

After all road signs are all explored completion, the activity of this cell of resetting.

The position cell is the variable relevant with the position only, and all it doesn't matter with movement velocity, direction of motion, defines this variable and is " position cell ", and is corresponding with biological position cell on function.In general dsp hardware, be presented as a variable.The reposition that in the process of exploring, occurs is with new neuron coding.In certain given position, a plurality of positions cellular activity, co.The density of position is relevant with robot residing environment position.Local fast in position angle change in location such as wall and doors, can acquire more terrestrial reference.After entire environment had all been learnt, environment can be covered by the position cell fully, and each position cell is corresponding with corresponding position.The position cell is the position that robot confirms oneself.

The road sign that is produced by said process forms " position cell " through Learning Algorithm.Movable and the common decision of road sign position angle fused cell (Fig. 3) by itself is provided in each position.If the exact position of robot cellular expression in the position, its movable maximum.When from then on robot removes the position, the activity of position cell is along with the distance of removing reduces gradually.

Each position cell all interconnects with all road sign position angle fused cells.Movable activity vector by road sign position angle fused cell carries out scalar product with corresponding connection weight vector.Therefore, the activity of position cell is by partial view of having learnt and current partial view decision.

Wherein $W_j=\underset{\mathrm{Kl}}{\overset{N_{\mathrm{PS}}}{\mathrm{\Σ}}}W.$

The learning process of position cell is observed the Hebbian rule.When robot is in the new environment, produce the new new new position of neuron coding automatically.This need not extraneous the intervention automatically.When the given threshold value of learning before in cellular activity region, position, new neuron can be participated in the new position of encoding automatically.

Above-mentioned road sign itself is a general notion; Here use " partial view around the unique point " this amount that clearly defines with " road sign " quantification; The variable that characterizes " road sign " is called the road sign cell; Partial view around the unique point representes that with road sign neuron k road sign neuron k can obtain through following formula:

ΔW＝I(t)□R

Wherein Δ W is from pixel i, and j is to the connection weight of k road sign, initial value be made as 0.I (t) constantly the pixel of t (coordinate i j) leaves the distance of unique point.R characterizes the coding whether this neuron has participated in this partial view, and the R value is that 0 or 1,0 expression connection weight is 0, does not participate in the coding of this partial view, and the coding of this partial view has been participated in 1 expression, and connection weight is I (t).

The movable X of k road sign cell ^Land(t), obtain by following formula

$X^{\mathrm{Land}}\left(t\right)=f\left(\frac{1}{\mathrm{NM}}\underset{i,j}{\overset{N,M}{\mathrm{\Σ}}}\right||\mathrm{\ΔW}\left(t\right)-I\left(t\right)|\left|\right)---\left(2\right)$

N and M are to be respectively horizontal ordinate and the ordinate number of pixels in the partial view.The definition of f is following

$f=\frac{1}{\mathrm{Thr}}{[x-\mathrm{Thr}]}^{+}---\left(3\right)$

Wherein Thr is a recognition threshold.When x>=0, [x] ⁺=x, otherwise be 0.

Should be understood that, concerning those of ordinary skills, can improve or conversion, and all these improvement and conversion all should belong to the protection domain of accompanying claims of the present invention according to above-mentioned explanation.

Claims (2)

1. a position cell bio-robot navigation algorithm is characterized in that, may further comprise the steps:

A1, at first robot wide-angle imaging head picked-up distant view photograph is translated into gradient map, uses the difference of gaussian wave filter to its filtering subsequently, detects unique point; Near to the unique point regional area is done log-polar transform, and these unique points are corresponding to specific road sign;

A2, the angle of obtaining with respect to direct north for each road sign is the position angle, direct north is provided by compass; The position cell of current time is confirmed on position angle and mark road jointly;

A3, the position cell of current time and the position cell in the previous moment confirm transition matrix jointly; In the time of in robot is explored in circumstances not known, constantly form transition matrix, transition matrix forms cognitive map;

A4, transition matrix and cognitive map determine the motion converter matrix jointly, the motion converter matrix is responsible for motion command is sent.

2. position according to claim 1 cell bio-robot navigation algorithm; It is characterized in that; Said road sign adopts the partial view quantification around the unique point; The variable that characterizes road sign is called the road sign cell, and the partial view around the unique point representes that with road sign neuron k road sign neuron k can obtain through following formula:

ΔW＝I(t)□R

Wherein Δ W is from pixel i, and j is to the connection weight of k road sign, initial value be made as 0.I (t) constantly the pixel of t (coordinate i j) leaves the distance of unique point; R characterizes the coding whether this neuron has participated in this partial view, and the R value is that 0 or 1,0 expression connection weight is 0, does not participate in the coding of this partial view, and the coding of this partial view has been participated in 1 expression, and connection weight is I (t);

The movable X of k road sign cell ^Land(t), obtain by following formula

$X^{\mathrm{Land}}\left(t\right)=f\left(\frac{1}{\mathrm{NM}}\underset{i,j}{\overset{N,M}{\mathrm{\Σ}}}\right||\mathrm{\ΔW}\left(t\right)-I\left(t\right)|\left|\right)$

N and M are to be respectively horizontal ordinate and the ordinate number of pixels in the partial view; The definition of f is following:

$f=\frac{1}{\mathrm{Thr}}{[x-\mathrm{Thr}]}^{+}$

Wherein Thr is a recognition threshold.When x>=0, [x] ⁺=x, otherwise be 0.

Images