CN107063260B - Bionic navigation method based on rat brain hippocampus structure cognitive map - Google Patents
Bionic navigation method based on rat brain hippocampus structure cognitive map Download PDFInfo
- Publication number
- CN107063260B CN107063260B CN201710180995.8A CN201710180995A CN107063260B CN 107063260 B CN107063260 B CN 107063260B CN 201710180995 A CN201710180995 A CN 201710180995A CN 107063260 B CN107063260 B CN 107063260B
- Authority
- CN
- China
- Prior art keywords
- rat
- cell
- grid
- cells
- environment
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/20—Instruments for performing navigational calculations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
Abstract
The invention discloses a bionic navigation method based on a rat brain hippocampus structure cognitive map, and belongs to the technical field of bionics. Firstly, a grid cell grid field model is constructed based on the stripe cells, secondly, a single-position cell position field model is constructed based on the grid cells, and finally, a cognitive map in the brain of the rat is constructed. On the basis of the constructed cognitive map, a feedforward neural network model comprising an input layer, a position cell layer, an action cell layer and an output layer is constructed, and a Q learning algorithm is adopted to realize a navigation task of a rat facing a certain target in a space environment. The method can be widely applied to the fields of bionic robot navigation, artificial intelligence and the like.
Description
Technical Field
The invention relates to a bionic navigation method based on a rat brain hippocampus structure cognitive map, and belongs to the technical field of bionics.
Background
The autonomous positioning and navigation ability of a certain target facing the space environment are of great importance for animals and autonomous mobile robots. Although the robot can determine the current position information of the robot through some priori information in a specific sensor or environment, the animal and the human can always quickly locate the current position information of the robot through incomplete spatial information from sense organs without any a priori knowledge. With the successive discovery of streak cells, grid cells and position cells related to environmental cognition in the brain of a rat, the possibility of deeply understanding the self-localization process of animals and human beings in the spatial environment is provided.
The three-way navigation experiment conducted by Tolman in 1948 shows that the rat can navigate through an internal 'cognitive map', and the cognitive map is the global expression of the spatial environment of the rat in the brain. The "cognitive map" plays a crucial role in spatial navigation in rats. Therefore, how to form the "cognitive map" and how to navigate in space based on the "cognitive map" is still an important experimental and theoretical problem.
In 1971, O' keefe and Dostrovesky first discovered positional cellular neurons with spatially localized activity in the hippocampal region of the hippocampal structure in rats. When a rat moves in a two-dimensional space, when the rat passes through a certain narrow area in the environment, cells at corresponding positions can discharge, and cells at single positions correspond to a cell activation domain at single positions, and the area is called as a cell position field. With further research, the single position field of the position cell is proved to be a basic component unit of the cognitive map, the position field of the single position cell has an accurate corresponding relation with the position of the rat, and a plurality of position cells jointly code the environment of the rat to form the cognitive map in the brain of the rat. In 2005, Hafting et al discovered another latticed cell neuron with strong spatial discharge characteristics in the entorhinal cortex of rats, which exhibited repetitive regular discharges in a specific region in space, referred to as the latticed field. Different from the position cells, the grid cells can discharge at a plurality of positions, grid fields of the grid cells are mutually overlapped to form grid nodes, and a regular triangle formed by connecting the grid nodes is distributed in the whole space environment of the rat. The hippocampal structure information pathway shows that the projection of entorhinal cortex to hippocampal fibers shows that the entorhinal cortex is the main input source of the hippocampus in the hippocampal structure, namely, grid cells are used as input to obtain the cell position field of a single position in the hippocampus. Krupic in 2012 equates to cell discharge sampling in the subnatal support and the hypodermis of the entorhinal cortex, indicating that there are periodic striped cell discharge field cells, called striped cells, which mainly integrate rat self-movement information (speed and direction information) to obtain striped cell discharge characteristics, i.e. striped cell stripe ripples. Fourier analysis shows that a plurality of stripe cell interactions can form a periodic grid cell grid field.
The single-position cell position field is a basic construction unit of a cognitive map, and how to obtain the single-position cell position field is an important theoretical problem. Since the discovery of the position cell in 1971, a plurality of different position cell position field models are proposed in succession, and since the discovery of the position cell, a plurality of different position cell position field models are proposed in succession, wherein the models comprise a Gaussian function model, a competition learning model, an independent component analysis model, a self-organizing map model, a Kalman filtering model and the like. However, the above models have a problem that they are modeled only for the position cells, and do not take into account the stripe cells and the grid cells which are related to the position cells.
When the rat enters a certain space environment, as the rat moves freely, the cell 'position field' is rapidly generated and covers the whole space environment where the rat is located. Namely, along with the continuous exploration of the environment by the rat, the position cell 'position field' forms a cognitive map representing the space environment of the rat. Some experimental studies have conducted relevant studies on rodent target-oriented navigation learning, and target-oriented navigation based on a hippocampal "cognitive map" is mainly accomplished by reinforcement learning, and biological studies have found that a cognitive map (site cell discharge activity) alone cannot correctly predict a rat's future movement direction, a Ventral Tegmental Area (VTA) of a brain is mainly dopaminergic neurons (dopaminergic neurons) associated with reward prediction error signals, and the dopaminergic neurons project information to Nucleus Accumbens (NA), and the nucleus accumbens input is mainly from hippocampus, and bi-directional fiber projection exists between prefrontal cortex and nucleus accumbens. Namely, the nucleus accumbens receives the spatial environment information of the rat from the hippocampus, receives the related reward prediction error information from the ventral tegmental area of the brain, and interacts with the prefrontal cortex to correctly predict the future movement direction of the rat. The prefrontal cortex of the brain is mainly neuron-action cells associated with movement. While the hippocampus is primarily the positional cell. Based on the biological correlation findings, the neural basis of the target-oriented navigation task of the rat may be synaptic regulation between cells at the hippocampus and nucleus accumbens related to reward signals, and the nucleus accumbens further projects information to the prefrontal cortex of the rat to realize correct prediction of the future movement direction of the rat. The rat target-oriented navigation task is realized through reinforcement learning between a continuous state space and an action space, wherein the continuous state space refers to a position cell discharge activity, namely a cognitive map in the rat brain. Based on the method, a feedforward neural network model comprising an input layer, a position cell layer, an action cell layer and an output layer is constructed, and a Q learning algorithm is adopted to realize a target-oriented navigation task of the rat.
Disclosure of Invention
In summary, the present invention is directed to a method for constructing a rat brain "cognitive map" based on striped cells, grid cells, and location cells. Secondly, on the basis of a cognitive map, a feedforward neural network model comprising an input layer, a position cell layer, an action cell layer and an output layer is constructed, and a Q learning algorithm is adopted to realize a target-oriented navigation task of the rat.
In order to achieve the purpose, the technical scheme adopted by the invention is a bionic navigation method based on a rat brain hippocampus structure cognitive map, and the overall model schematic diagram is shown in figure 1. According to the overall model diagram, the model consists of two modules. The first module is how to obtain the expression of the rat brain to the spatial environment, namely a cognitive map, based on the rat autoromotor information, and the structural diagram of the cognitive map is shown in fig. 2. The second module is to realize a certain target navigation task facing the space environment of the rat through Q learning on the basis of a cognitive map, and the structural schematic diagram of the second module is shown in fig. 3.
The two models are realized by adopting the following technical scheme:
s1 a virtual rat is constructed to randomly explore the two-dimensional space environment and obtain the two-dimensional space motion trail diagram, wherein the self-motion information of the rat is composed of speed and direction information, the horizontal axis is used as the reference direction, alphatRepresents the current head orientation of the rat, vtRepresenting the current rat velocity.
S2 obtains a striped cell space characterization, i.e., striped cell strip ripple, based on rat autoromotor information, i.e., velocity and direction information. Striped cell discharge activity
Comprises the following steps:
wherein, r ═ (x, y) represents the coordinate of the current environment position of the rat, and kiThe representative wave vector, i is 1,2,3, the direction of the wave vector represents the direction of the wave with equal phase, and the magnitude is called wave number ki，kiComprises the following steps:
where λ cos represents the cos wave wavelength.
S3 obtains a lattice cell spatial representation, i.e., a lattice cell lattice field, based on the fringe cell spatial representation. Superposing three striped cells with 60-degree difference in orientation to obtain a grid cell field, wherein the grid cell field spatial characterization psi (r) is as follows:
wherein, r ═ (x, y) represents the coordinates of the current environment position of the rat.
S4 deriving a positional cell space representation based on the grid cell space representation, i.e. a single positional cell position field, the positional cell space representation P (x, y) being:
wherein, WnAnd the connection weight value between the nth grid cell and the position cell is represented, gn (x, y) represents the activation rate of the nth grid cell at the position point of the space environment (x, y), N represents the number of grid cells, and N is 4,10 and 20.
In the process of continuously exploring the environment, the S5 rat forms the cell discharge characteristics of the position at each position point, and finally forms the expression of the spatial environment in the rat brain, namely a cognitive map.
S6, on the basis of the cognitive map, a feedforward neural network model consisting of an input layer, position cells, action cells and an output layer is built, and a rat target-oriented navigation task is realized through a Q learning algorithm.
Drawings
FIG. 1 is a schematic view of the model of the present invention as a whole.
FIG. 2 is a schematic diagram of a rat brain hippocampal cognitive map construction structure.
Fig. 3 is a schematic diagram of a target-oriented navigation structure based on a rat brain hippocampal cognitive map and Q learning.
FIG. 4 is a schematic diagram of spatial motion trajectory of rat.
FIG. 5 is a schematic diagram of the field of the striped cell stripe ripple grid cell grid
FIG. 6 is a graph showing the relationship between the grid field interval λ of the grid cells and the wavelength λ cos of the two-dimensional cos wave.
FIG. 7 is a schematic diagram of the results of the grid cell linear superposition experiment without Sigmoid function processing
Figure 8Sigmoid function diagram
FIG. 9 is a schematic diagram of the experimental results of grid cell linear superposition processed by Sigmoid function
FIG. 10 cognitive mapping process schematic
FIG. 11 is a schematic diagram of an experimental environment
FIG. 12 is a schematic diagram of a feedforward neural network model.
FIG. 13 is a schematic diagram of a feed-forward network of the input layer to the site cell layer.
FIG. 14 is a schematic representation of spatial navigation of a rat.
FIG. 15 is a schematic diagram of a feedforward network model constructed from an input layer (position cells) and action cells.
FIG. 16 is a schematic diagram of the experimental results of 40 runs of rats
FIG. 17 is a graph showing the number of steps required for a rat to reach a target position
Detailed Description
The invention is further explained below with reference to the figures and examples.
S1 rat autoromotive information consists of head orientation and velocity information. With the horizontal axis as reference, αtRepresenting the current head orientation of the rat. v. oftRepresenting the current rat velocity. Δ t represents a time period. Based on the current autokinetic information of the rat and the position information (x) of the rat at the last momentt-1,yt-1) To calculate the current position information (x) of the ratt,yt) The current self-motion information refers to the head orientation αtAnd velocity vtAs shown in formula (5).
The coordinate of the initial position of the rat is (x)0,y0) The spatial movement locus diagram of the rat is shown in fig. 4 (0, 0).
S2 striped cells are found in the 3 rd layer of the entorhinal cortex, the discharge activity of the cells in the two-dimensional space environment is a cluster of stripe ripples, the striped cells integrate the rat' S self-movement information and then transmit the information to the 2 nd layer of the entorhinal cortex, and the grid cell grid field with different spatial phases, orientations and spacings formed by the generated cluster ripples through superposition is shown in fig. 5 based on the schematic diagram of the grid cell grid field of the striped cell stripe ripples.
For the striated cell discharge activity, a two-dimensional cos wave is used as shown in formula (6).
Where, r ═ (x, y) represents the position coordinates of the rat's current environment, the wave vector direction represents the direction of wave equiphase travel, and the magnitude is called wave number kiAs shown in formula (7).
Where λ cos represents the cos wave wavelength.
S3 biological research proves that the grid cell field is in regular triangle shape and is distributed in the whole environment of the rat. Based on this, the grid cell activation rate function can be represented by the superposition of the discharge activities of three striped cells, whose wave vectors are oriented at a phase difference of 60 °, as shown in equation (8).
As shown in equation (8), when r is (0,0), Ψ (r) has a maximum value of 1. If an arbitrary spatial phase r0 in the spatial environment is selected as (x0, y0) as a certain peak point of the grid field of the grid cell, the grid cell activation rate function is converted to the formula (9).
ψ(r)＝ψ(r-r0) (9)
The wave vector is a function of the wave number, and the grid field spacing of the grid cells is taken as 1 parameter for characterizing the spatial discharge characteristics of the grid cells. As shown in fig. 6, circles represent grid cell field nodes, horizontal stripes represent striped cell two-dimensional cos waves, and the relationship between grid cell field distance λ and the wavelength λ cos of the two-dimensional cos waves is shown in formula (10).
Further, from the equation (7), the relationship between the wave number and the field distance λ of the grid cell is shown in the equation (11).
S4 in order to obtain a regular triangle grid cell field consistent with the biological finding, three stripe cells with wave vector orientation difference of 60 degrees are selected and superposed to obtain the required grid cell field. 60 degrees, 120 degrees and 180 degrees, and k1, k2 and k3 are selected as shown in formula (12).
Where θ represents the grid field orientation of the grid cell.
From equation (8), psi (r) is between [ -1/2, 1], and in order to make the grid cell activation rate between 0 and 1, the grid cell activation rate function is transformed as shown in equation (13).
The grid cell activation rate function obtained by substituting the formula (8) and the formula (9) into the formula (13) is shown as the formula (14).
Biological studies at S5 found that projections of entorhinal cortex into hippocampal fibers in hippocampal structures suggested that entorhinal cortex was the major input source to hippocampus in hippocampal structures. Both the grid cells and the site cells are nerve cells, which are composed of cell bodies and cell processes, wherein the cell processes are elongated parts extending from the cell bodies, and the elongated parts are divided into dendrites and axons. Each neuron has only one axon, transmits signals to other tissues or to another neuron, has a plurality of dendrites, receives stimulation and transmits excitation into cells, and so on between grid cells and position cells. Namely, the position cell receives information from grid cells with different spatial characteristics, and then the information and the weight between grid cells connected with the position cell are weighted and summed to obtain the discharge characteristic of the position cell, wherein the function of the connection weight between the grid cells and the position cell is shown as a formula (15).
Wherein, Wn represents the connection weight between the nth grid cell and the position cell, λ n represents the grid field distance of the nth grid cell, and σ (σ ═ 8cm) represents the standard deviation of the discharge activation domain of the position cell.
S6 shows the positional cell activation rate as shown in equation (16) from equations (14) and (15).
Wherein, Wn represents the connection weight between the nth grid cell and the position cell, gn (x, y) represents the activation rate of the nth grid cell at the position point of the space environment (x, y), N represents the number of grid cells, and N is 4,10, 20.
S7 biological research shows that the output obtained by simply linearly superposing grid cell input and connection weights between grid cells and position cells is a position field with a plurality of activation domains, and the result is schematically shown in FIG. 7, which is inconsistent with the conclusion that a single position cell corresponds to a single position field as proved by related research. According to the method, a Sigmoid function is introduced on the basis of grid field input of grid cells and weighted summation of connection weights between grid cells and position cells, a Sigmoid function graph is shown in fig. 8, output after linear superposition is processed, a single position cell position field consistent with biological research findings is obtained, the mapping relation between the grid cells and the single position cell position field is realized, an experimental result schematic diagram is shown in fig. 9, and the position cell activation rate after being processed by the Sigmoid function is shown in a formula (17).
P'(x,y)＝1/(1+e-(P-b)/a) (17)
Wherein, P represents the cell activation rate of the position, a represents the tilting coefficient of the Sigmoid function, and b represents the center of the Sigmoid function.
In the process of continuously exploring the environment, the rat of S8 forms the cell discharge characteristics of the position at each position point, and finally forms the expression of the spatial environment in the brain of the rat, namely a cognitive map, and the forming process schematic diagram of the cognitive map is shown in figure 10.
S9 some experimental studies have conducted relevant studies on rodent target-oriented navigation learning, and target-oriented navigation based on hippocampal "cognitive map" is mainly accomplished by reinforcement learning, and biological studies have found that cognitive map (location cell discharge activity) alone cannot correctly predict the rat' S future movement direction, Ventral Tegmental Area (VTA) of brain is mainly dopaminergic neurons (dopaminergicneurons) associated with reward prediction error signals, and that information is further projected by dopaminergic neurons to Nucleus Accumbens (NA), and nucleus accumbens input is mainly from hippocampus, and bi-directional fiber projection exists between prefrontal cortex and nucleus accumbens. Namely, the nucleus accumbens receives the spatial environment information of the rat from the hippocampus, receives the related reward prediction error information from the ventral tegmental area of the brain, and interacts with the prefrontal cortex to correctly predict the future movement direction of the rat. The prefrontal cortex of the brain is mainly neuron-action cells associated with movement. While the hippocampus is primarily the positional cell. Based on the biological correlation findings, the neural basis of the target-oriented navigation task of the rat may be synaptic regulation between cells at the hippocampus and nucleus accumbens related to reward signals, and the nucleus accumbens further projects information to the prefrontal cortex of the rat to realize correct prediction of the future movement direction of the rat. The rat target-oriented navigation task is realized through reinforcement learning between a continuous state space and an action space, wherein the continuous state space refers to a position cell discharge activity, namely a cognitive map in the rat brain. Based on the method, a feedforward neural network model comprising an input layer, a position cell layer, an action cell layer and an output layer is constructed, and a Q learning algorithm is adopted to realize a target-oriented navigation task of the rat.
The S10 experimental environment is a square box (as shown in fig. 11) with dimensions of 10000 × 10000 dots. As the rat continuously explores the space environment of the rat, the position cell position fields of all position points in the space are gradually formed, and finally, the internal map representation-cognitive map of the environment is formed in the rat brain. Therefore, in the model, the present position point information (x) of the rat is shown in the textt,yt) As input information.
S11, a feedforward neural network model composed of an input layer, position cells, action cells and an output layer is constructed to realize the rat target-oriented navigation task, and the feedforward neural network model is shown in figure 12.
S12 inputting layer to positionThe cell layer feed forward network is shown in figure 13. At the input layer, input X: (x)t,yt) Information is entered for the current position of the rat. The feedforward network is a fully connected network, and each neuron of the input layer passes through a connection weight Wi＝[wi,1,wi,2···wi,n]And all the neurons are connected with the output layer of the feedforward network in sequence. Here, i is 1 … Q, and Q is 500, the total number of the site cells. The weight being given by the function fuFor random initialization, function fuDescribed by the following formula (18).
In the formula, u is the value derived from obeying [ 0; 1], v ═ 0.5 and σ ═ 0.2. The positional cell discharge rate is calculated using the input information and the weights (see equation (19)), and the weights are first randomly initialized. With a competitive learning algorithm, the positional cell is excited by a particular input, thereby making the positional cell selective to spatial position.
The cell discharge rate at the ith position is described by the following equation (19):
in the formula, σfThe width of the positional cell position field is defined as 0.07, n is the spatial dimension of the input information, and the norm represents the euclidean distance. The weight value in the constructed feedforward neural network model is adjusted according to the mechanism that the winner is the king, namely the cell neuron chi at the position winning by adopting the competitive learning algorithmtThe weight value between the input information and the cell neuron X at the winning position can be changed, and the rest are not changedtDescribed by the following equation (20):
χt＝argmini||Xt-Wt i|| (20)
the weight of the winning neuron is changed according to the following formula (21):
in the formula, 0 < alpha < 1 represents a learning efficiency factor.
S13 is that the current position information of rat is used as input to affect the discharge activity of position cell, and the position cell is connected with motor neuron to generate certain action through Q learning. The rat learns the navigation from any starting position to the target position by continuous learning, and the spatial navigation is schematically shown in fig. 14. Using the experimental environment shown in fig. 11, the rat was in an environment with a starting position at the lower left of the experimental environment (as indicated by the dotted circle in fig. 14). The target point is located at the upper right of the experimental environment (as shown by the square in fig. 14). Initially, the rat randomly explores the environment, finds the target location point in the process of randomly exploring the environment (the random exploration path is shown by the dotted line in fig. 14), and after the rat learns for a while, it can quickly find the shortest path from the starting location to the target location. During the experiment, each time the rat finds the target position point, the rat is placed again at the initial position to start a new round of experiment. In the constructed model, in most cases (80%), rats were able to find the target location point within 200 steps in the first experiment, that is, 200 steps were sufficient for rats to find the target location point even when the rats were first randomly exploring the experimental environment herein.
S14 a reinforcement learning algorithm is commonly used based on hippocampal navigation studies. Used herein is a Q learning algorithm similar to Reynolds. The algorithm is applied to a two-layer feedforward neural network (as shown in fig. 15) with the location cells as the input to the network. Each position cell is connected in turn to motor neurons representing 8 different directions (north (N), Northeast (NE), east (E), Southeast (SE), south (S), Southwest (SE), west (W), Northwest (NW)) respectively. The actual direction of motion is determined by the maximum Q value of the eight possible directions. The horizontal rat movement to the west to the east is described by the following equations (22) and (23):
Δx＝±(Δs+c·ψx) (22)
Δy＝c·ψy (23)
in the formula, Δ s-500 represents the step size per step of the rat, ψxAnd psiyFrom obey [ -1; 1]Uniformly distributed random values, c 100 is the noise amplitude. The minus sign indicates the rat moves westward and the plus sign indicates the rat moves eastward. Likewise, the movements for the southwest and northeast rats are described by the following equations (24) and (25):
s15 when the rat moves to the current position, the Q value calculated is 0, the rat is not limited to 8 directions (north (N), Northeast (NE), east (E), Southeast (SE), south (S), Southwest (SE), west (W) and Northwest (NW)) at the current position, but carries out random exploration movement, and the movement direction is uncertain. In this case, however, the rat has a possibility of keeping the direction unchanged at 1-pkAnd the probability of its random selection of a new direction is pk0.25. When the Q value is not 0, the rat will determine the moving direction at the next moment of the current position according to the Q value most of the time as the rat continuously explores the environment. The learning mechanism from the location cell to the action cell is the Q learning algorithm. For simplicity, the cell discharge rate at the ith position at time t is described by the following equation (26):
in the formula, i is 1 … Q, and Q is 500 is the total number of the site cells.
S16 defines the action value function by the following equation (27):
in the formula, gammai,aThe connection weight between the cell at the ith position and the motor neuron a is represented. The rule is learned using average Q as mentioned by Reynolds. That is, the action a actually generated at the time t is updated according to the following formula (28)tWeight of (2)
In the formula, β ═ 0.7 represents the learning rate, δ ═ 0.7 represents the reduction coefficient, and R represents the reward. Will reward function RtDescribed by the following function (29):
the invention mainly constructs a rat brain cognitive map based on stripe cells, grid cells and position cells. On the basis of a cognitive map, a feedforward neural network model comprising an input layer, a position cell layer, an action cell layer and an output layer is constructed, and a Q learning algorithm is adopted to realize a rat target-oriented navigation task. The experimental results of the 40-time running tracks of the rat and the number of steps required for the rat to reach the target position are respectively shown in fig. 17.
Claims (1)
1. A bionic navigation method based on rat brain hippocampal structure cognitive map comprises a model consisting of two modules; the first module is based on rat self-movement information, how to obtain the expression of the rat brain to the space environment, namely a cognitive map; the second module realizes a rat navigation task facing a certain target of the space environment through Q learning on the basis of a cognitive map;
the two models are realized by adopting the following technical scheme:
s1 a virtual rat is constructed to randomly explore the two-dimensional space environment and obtain the two-dimensional space motion trail diagram, wherein the self-motion information of the rat is composed of speed and direction information, the horizontal axis is used as the reference direction, alphatRepresents the current head orientation of the rat, vtRepresenting the current speed of the rat;
s2 obtaining stripe cell space representation based on rat self-movement information, namely speed and direction information, namely stripe cell strip ripple; striped cell discharge activity
Comprises the following steps:
wherein, r ═ (x, y) represents the coordinate of the current environment position of the rat, and kiThe representative wave vector, i is 1,2,3, the direction of the wave vector represents the direction of the wave with equal phase, and the magnitude is called wave number ki，kiComprises the following steps:
wherein λ cos represents a cos wave wavelength;
s3 obtaining grid cell space representation based on the stripe cell space representation, namely grid cell grid field; superposing three striped cells with 60-degree difference in orientation to obtain a grid cell field, wherein the grid cell field spatial characterization psi (r) is as follows:
wherein, r ═ (x, y) represents the coordinates of the current environment position of the rat;
s4 deriving a positional cell space representation based on the grid cell space representation, i.e. a single positional cell position field, the positional cell space representation P' (x, y) being:
P'(x,y)＝1/(1+e-(P-b)/a) (4)
wherein, a represents the tilting coefficient of the Sigmoid function, b represents the center of the Sigmoid function, and P represents the activation rate of the cells at the position, and the mathematical expression is as follows:
Wnrepresenting the connection weight value between the nth grid cell and the position cell, wherein the mathematical expression is as follows:
gn(x, y) represents the activation rate of the nth grid cell at the position point of the space environment (x, y), N represents the number of grid cells, and N is 4,10, 20; lambda [ alpha ]nThe field interval of the nth grid cell grid is obtained; sigma is the standard deviation of the position field of the position cell;
s5, in the process of continuously exploring the environment, the rat forms the cell discharge characteristics of the position at each position point, and finally forms the expression of the spatial environment in the rat brain, namely a cognitive map;
s6, on the basis of the cognitive map, a feedforward neural network model consisting of an input layer, position cells, action cells and an output layer is built, and a rat target-oriented navigation task is realized through a Q learning algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710180995.8A CN107063260B (en) | 2017-03-24 | 2017-03-24 | Bionic navigation method based on rat brain hippocampus structure cognitive map |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710180995.8A CN107063260B (en) | 2017-03-24 | 2017-03-24 | Bionic navigation method based on rat brain hippocampus structure cognitive map |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107063260A CN107063260A (en) | 2017-08-18 |
| CN107063260B true CN107063260B (en) | 2021-03-30 |
Family
ID=59618250
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710180995.8A Active CN107063260B (en) | 2017-03-24 | 2017-03-24 | Bionic navigation method based on rat brain hippocampus structure cognitive map |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107063260B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109425340B (en) * | 2017-08-28 | 2022-03-22 | 中国科学院沈阳自动化研究所 | Brain-like robot navigation method based on Bayes multi-mode perception fusion |
| CN108362284A (en) * | 2018-01-22 | 2018-08-03 | 北京工业大学 | A kind of air navigation aid based on bionical hippocampus cognitive map |
| CN109000655B (en) * | 2018-06-11 | 2021-11-26 | 东北师范大学 | Bionic indoor positioning and navigation method for robot |
| CN109668566B (en) * | 2018-12-05 | 2022-05-13 | 大连理工大学 | Robot scene cognition map construction and navigation method based on mouse brain positioning cells |
| CN109886384B (en) * | 2019-02-15 | 2021-01-05 | 北京工业大学 | Bionic navigation method based on mouse brain hippocampus grid cell reconstruction |
| CN110210462A (en) * | 2019-07-02 | 2019-09-06 | 北京工业大学 | A kind of bionical hippocampus cognitive map construction method based on convolutional neural networks |
| CN112525194B (en) * | 2020-10-28 | 2023-11-03 | 北京工业大学 | Cognitive navigation method based on in vivo source information and exogenous information of sea horse-striatum |
| CN113643748B (en) * | 2021-07-28 | 2024-03-12 | 哈尔滨工业大学 | Grid cell space information decoding method and system |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102401656A (en) * | 2011-11-08 | 2012-04-04 | 中国人民解放军第四军医大学 | Place cell bionic robot navigation algorithm |
| CN102799179B (en) * | 2012-07-06 | 2014-12-31 | 山东大学 | Mobile robot path planning algorithm based on single-chain sequential backtracking Q-learning |
| CN103699125B (en) * | 2013-12-09 | 2016-04-06 | 北京工业大学 | A kind of robot simulation air navigation aid based on the navigation of mouse cerebral hippocampal |
| CN103778332A (en) * | 2014-01-16 | 2014-05-07 | 北京工业大学 | Method for analyzing function of mouse hippocampus structure in environment cognition process |
| CN104571113B (en) * | 2015-01-20 | 2017-07-11 | 新智认知数据服务有限公司 | The paths planning method of mobile robot |
| CN106125730B (en) * | 2016-07-10 | 2019-04-30 | 北京工业大学 | A kind of robot navigation's map constructing method based on mouse cerebral hippocampal spatial cell |
-
2017
- 2017-03-24 CN CN201710180995.8A patent/CN107063260B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN107063260A (en) | 2017-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107063260B (en) | Bionic navigation method based on rat brain hippocampus structure cognitive map | |
| Patle et al. | A review: On path planning strategies for navigation of mobile robot | |
| CN109886384B (en) | Bionic navigation method based on mouse brain hippocampus grid cell reconstruction | |
| Praczyk | Neural anti-collision system for Autonomous Surface Vehicle | |
| Nikolos et al. | UAV path planning using evolutionary algorithms | |
| Precup et al. | Grey wolf optimizer-based approaches to path planning and fuzzy logic-based tracking control for mobile robots | |
| Ni et al. | An improved DSA-based approach for multi-AUV cooperative search | |
| Kozma et al. | Computational aspects of cognition and consciousness in intelligent devices | |
| Biswas et al. | A particle swarm optimization based path planning method for autonomous systems in unknown terrain | |
| Kurdi et al. | Proposed system of artificial Neural Network for positioning and navigation of UAV-UGV | |
| Pessin et al. | Intelligent control and evolutionary strategies applied to multirobotic systems | |
| Masár et al. | Modification of PSO algorithm for the purpose of space exploration | |
| Jiang et al. | Learning decentralized control policies for multi-robot formation | |
| Peng et al. | Pattern formation in constrained environments: A swarm robot target trapping method | |
| Heidari et al. | Improved black hole algorithm for efficient low observable UCAV path planning in constrained aerospace | |
| Guerrero-González et al. | A biologically inspired neural network for navigation with obstacle avoidance in autonomous underwater and surface vehicles | |
| Cuperlier et al. | Transition cells for navigation and planning in an unknown environment | |
| Meng et al. | Self-adaptive distributed multi-task allocation in a multi-robot system | |
| Huang et al. | Robotic organism targets regional coverage capture path planning for marine aquafarm based on value iteration network | |
| Sellers et al. | Deep learning-based heterogeneous system for autonomous navigation | |
| Rashid et al. | Polygon Shape Formation for Multi-Mobile Robots in a Global Knowledge Environment. | |
| Guoqing et al. | Bio-inspired autonomous navigation system for logistics mobile robots with inertial AHRS | |
| Asmaa et al. | UAV Control Architecture | |
| Makino et al. | Development of a motion planning system for an agricultural mobile robot | |
| Weirens et al. | Simulation toolset for localization and control of swarming vehicles using random finite set theory |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |