A brain-inspired navigation method integrating self-motion cues and external sensory information

Abstract

The present invention discloses a brain-inspired navigation method for fusing self-motion cues and external sensory information, comprising constructing a head-direction cell information fusion model using a continuous attractor neural network. The model encodes the yaw angle of self-motion cues and external sensory information, generating activity changes, which are converted into the discharge rate of a group of head-direction cells through an activation function; decoding the yaw angle of a drone using the discharge rate of the group of head-direction cells; constructing a three-dimensional periodic grid cell information fusion model using a continuous attractor neural network, encoding the decoded yaw angle, speed of self-motion cues, and position of external sensory information using the model, generating activity changes; and determining the drone's position by decoding the motion state of wave packets formed based on the activity of the three-dimensional periodic grid cell information fusion model. The present invention fuses the self-motion cues provided by an inertial navigation system with the external sensory information provided by a global navigation satellite system, providing an efficient and reliable brain-inspired navigation method for drones.

Description

Technical Field

[0001] The present invention relates to the field of intelligent navigation technology, and in particular to a brain-inspired navigation method that integrates self-motion clues and external perception information. Background Art

[0002] Drone technology has a wide range of applications and is increasingly integrated into numerous aspects of our daily lives. With the integration of artificial intelligence, drones' operational autonomy and situational awareness capabilities have been significantly enhanced. As a core component in achieving these missions, extensive research is underway into navigation systems to further enhance their autonomous navigation capabilities.

[0003] As scientists' understanding of the brain continues to deepen, the discovery and classification of neurons with unique spatial sensitivity are also progressing. These spatial representation-based cells provide new perspectives and theoretical foundations for the study of navigation systems, leading to increasingly in-depth and extensive exploration and application in this field. Currently, within the field of spatial representation cells, specific cell types such as head heading cells and grid cells play crucial roles in mammalian navigation. Mammalian navigation relies heavily on precise directional perception, a capability believed to be closely related to the mechanisms by which head heading cells process environmental information. These cells exhibit specific firing patterns when a rat's head is aligned with a specific direction, demonstrating their unique spatial selectivity. Unlike cells that focus on head direction, grid cells produce a distinctive hexagonal firing pattern when a rat moves freely in two-dimensional space, providing another sophisticated means of encoding spatial orientation. Notably, many animals, such as bats, need to navigate and localize themselves in more complex three-dimensional spaces. The demonstrated ability of grid cells in these animals to encode three-dimensional space is of great significance in neuroscience research.

[0004] The discovery of spatial representation cells not only enriches our understanding of the brain's navigation mechanisms but also provides a solid foundation for further exploration of the principles and methods of neural coding. These findings have potentially significant implications for the development and application of future navigation technologies, particularly in artificial intelligence and robotics. However, most existing brain-inspired navigation models rely on vision as their primary information source. To enhance compatibility with drones equipped with general-purpose sensors, the development of a brain-inspired navigation model remains a challenge. Summary of the Invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a brain-inspired navigation method that can integrate the self-motion clues provided by the inertial navigation system in the drone and the external perception information provided by the global navigation satellite system to improve the positioning accuracy of the drone.

[0006] Technical solution: To achieve the above objectives, the present invention provides a brain-inspired navigation method that integrates self-motion cues and external sensory information, comprising the following steps:

[0007] Step 1: The drone's navigation system includes an inertial navigation system and a global navigation satellite system. The inertial navigation system provides egomotion cues, and the global navigation satellite system provides external perception information. The egomotion cues include yaw angle and speed, and the external perception information includes yaw angle and position information.

[0008] Step 2: A continuous attractor neural network was used to construct a head-facing cell information fusion model to simulate the discharge characteristics of head-facing cells. Specifically, the model encoded the yaw angle of self-motion cues and external sensory information to generate activity changes, which were converted into the group discharge rate of head-facing cells through an activation function.

[0009] Step 3: Use the group firing rate of the head-facing cells to decode the yaw angle of the drone. The yaw angle of the drone is the fusion result of the yaw angle of the self-motion information and the yaw angle of the external perception information;

[0010] Step 4: Use a continuous attractor neural network to construct a three-dimensional periodic grid cell information fusion model to simulate the discharge characteristics of three-dimensional periodic grid cells. That is, use this model to encode the yaw angle, speed of self-motion cues, and position of external sensory information decoded in step 3 to produce activity changes;

[0011] Step 5: Based on the wave packet formed by the activity of the three-dimensional periodic grid cell information fusion model, the position of the UAV is determined by decoding the motion state of the wave packet.

[0012] Among them, during the encoding process described in step 2, the activity of the head-facing cell information fusion model Updates as follows:

[0013]

[0014] in, is the excitatory connection weight between head-facing cells, w INH,HD For local inhibition, represents global suppression, represents the firing rate of the head-facing cell, I m HD represents the input of the information fusion model of the head facing the cell, λ HD,w is the weight coefficient, τ HD =1 is the time constant; m is the head facing the cell at the current moment, and n is the head facing the cell at the previous moment;

[0015] The update rule for the excitatory connection weights between head-facing cells is:

[0016]

[0017] in, is the discharge rate used for learning at time t, k HD =0.1 is the adjustment coefficient; is the discharge rate used for learning at time t-1;

[0018] The calculation method is as follows:

[0019]

[0020] Among them, σ HD is the variance, Indicates the current yaw angle of the drone ψ HD with the head toward the cell's preferred direction The difference between The calculation is as follows:

[0021]

[0022] Self-motion information and external perception information serve as input to head heading cells I m HD :

[0023] I m HD (t) = I m S,ψ (t)+I m E,ψ (t),

[0024] Among them, I m S,ψ is the yaw angle input of the self-motion cue, I m E,ψ The yaw angle input is the external perception information.

[0025] The head orientation cell information fusion model encodes the yaw angle provided by the self-motion clues and the external perception information based on the Gaussian function. The encoding calculation method is as follows:

[0026]

[0027] Among them, σ *,ψ is the variance, ψ * The yaw angle provided by the self-motion clue or the yaw angle provided by the external sensory information; indicates the preferred direction of the head toward the cell; Indicates Im S,ψ (t) or I m E,ψ (t).

[0028] Wherein, the discharge rate Calculate using the hyperbolic tangent function:

[0029]

[0030] The decoding in step 3 is to calculate the yaw angle ψ by using the group vector method to decode the group discharge rate of the cell toward which the head is facing. HD , the process is:

[0031]

[0032]

[0033]

[0034] in, represents the firing rate of the head-direction cell.

[0035] In the encoding process described in step 4, the activity of the three-dimensional periodic grid cell information fusion model is updated as follows:

[0036]

[0037] in, Indicates local excitatory activity, Indicates activity inhibition, ΔP GC represents the activity offset, Represents the input of external perception information; i, j, k represent the three-dimensional cells in the three-dimensional periodic grid cell information fusion model.

[0038] Wherein, the local excitatory activity Excitatory connection weights Build, the build process is as follows:

[0039]

[0040] in, and is the dimension of the three-dimensional periodic grid cell information fusion model;

[0041] Excitatory connection weights The calculation process is as follows:

[0042]

[0043] in, and represents the variance, and Represents three-dimensional periodic grid cell sequences and The position x corresponding to the UAV in the three-dimensional periodic grid cell information fusion model c ,y c and z c the distance between them;

[0044] and The calculation process is as follows:

[0045]

[0046] Wherein, the inhibitory activity The calculation process is as follows:

[0047]

[0048] in, To suppress the weight, Indicates global inhibition; is the dimension of the three-dimensional periodic grid cell information fusion model.

[0049] The activity shift It is used to encode the velocity provided by the self-motion clues and the decoded yaw angle. The encoding process is as follows:

[0050]

[0051] Among them, δ x0 , δ y0 , δ z0 is the integer part of the remainder γ, which is the horizontal velocity v obtained by the self-motion clue S , height and speed and yaw angle ψ HD Determined together, the calculation process is as follows:

[0052]

[0053]

[0054]

[0055]

[0056] in, and is the adjustment coefficient; δ xf , δ yf , δ zfis the decimal part of the remainder; a, b are variables in the function f(a,b);

[0057] The input of the external perception information It is used to correct the accumulated error of the three-dimensional periodic grid cell information fusion model. The three-dimensional periodic grid cell information fusion model encodes the position of the external perception information based on the Gaussian function. The encoding process is as follows:

[0058]

[0059]

[0060] Among them, the mod() function is used to calculate the position x provided by the external perception information E 、y E and z E With parameters and the remainder of and Represents a cyclic sequence based on the positions provided by external sensory information.

[0061] The method for determining the position of the UAV by decoding the motion state of the wave packet in step 5 is:

[0062] Step 501: Calculate the center position of the wave packet (x GC ,y GC ,z GC )as follows:

[0063]

[0064] in, is the dimension of the three-dimensional periodic grid cell information fusion model; is a three-dimensional periodic grid cell sequence;

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071] k size =2π / n GC ;

[0072] Among them, n GC For the same value or or

[0073] Step 502: According to the change of the position of the wave packet, the wave packet movement speed is obtained as follows:

[0074]

[0075] Step 503: The position (x, y, z) of the UAV is further obtained according to the velocity of the wave packet as follows:

[0076]

[0077] Among them, c x , c y and c z is the adjustment coefficient.

[0078] Beneficial effects: The present invention has the following advantages: 1. The present invention integrates the self-motion clues provided by the inertial navigation system and the external perception information provided by the global navigation satellite system, providing an efficient and reliable positioning and navigation method for drones;

[0079] 2. This invention uses a constructed head heading cell information fusion model to fuse self-motion cues and external sensory information to generate discharge characteristics for the yaw angle. It further uses the discharge rate of head heading cells to decode the yaw angle of the drone, enabling the drone to more accurately determine its own yaw angle and improving the accuracy of brain-inspired navigation.

[0080] 3. The present invention utilizes a constructed three-dimensional periodic grid cell information fusion model to fuse self-motion cues and external perception information to generate discharge characteristics. At the same time, the input of external perception information can correct the accumulated error of the three-dimensional periodic grid cell model, further obtain the wave packet formed by cell activity, and determine the position of the UAV by decoding the motion state of the wave packet. This can effectively determine the spatial position of the UAV, thereby improving the reliability and real-time performance of brain-inspired navigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 It is a schematic flow chart of the present invention;

[0082] Figure 2 This is a comparison chart between the brain-inspired positioning result and the true value after correction of position information without external sensory information;

[0083] Figure 3 This is a comparison chart between the brain-inspired positioning results of the fusion of self-motion clues and external perception information and the true value. DETAILED DESCRIPTION

[0084] The technical solution of the present invention is described in detail below with reference to the embodiments and drawings.

[0085] like Figure 1 As shown, a brain-inspired navigation method that integrates self-motion cues and external perception information includes the following steps:

[0086] Step 1: The drone's navigation system includes an inertial navigation system and a global navigation satellite system. The inertial navigation system provides egomotion cues, and the global navigation satellite system provides external perception information. The egomotion cues include yaw angle and speed, and the external perception information includes yaw angle and position information.

[0087] Step 2: A continuous attractor neural network was used to construct a head-facing cell information fusion model to simulate the discharge characteristics of head-facing cells. Specifically, the model encoded the yaw angle of self-motion cues and external sensory information to generate activity changes, which were converted into the group discharge rate of head-facing cells through an activation function.

[0088] Step 3: Use the group firing rate of the head-facing cells to decode the yaw angle of the drone. The yaw angle of the drone is the fusion result of the yaw angle of the self-motion information and the yaw angle of the external perception information;

[0089] Step 4: Use a continuous attractor neural network to construct a three-dimensional periodic grid cell information fusion model to simulate the discharge characteristics of three-dimensional periodic grid cells. That is, use this model to encode the yaw angle, speed of self-motion cues, and position of external sensory information decoded in step 3 to produce activity changes;

[0090] Step 5: Based on the wave packet formed by the activity of the three-dimensional periodic grid cell information fusion model, the position of the UAV is determined by decoding the motion state of the wave packet.

[0091] In step 2, the head heading cell model is a one-dimensional model constructed based on a continuous attractor neural network, and in step 4, the grid cell model is a three-dimensional model constructed based on a continuous attractor neural network.

[0092] Among them, during the encoding process described in step 2, the activity of the head-facing cell information fusion model Updates as follows:

[0093]

[0094] in, is the excitatory connection weight between head-facing cells, w INH,HD For local inhibition, represents global suppression, represents the firing rate of the head-facing cell, I m HDrepresents the input of the information fusion model of the head facing the cell, λ HD,w is the weight coefficient, τ HD =1 is the time constant; m is the head facing the cell at the current moment, and n is the head facing the cell at the previous moment;

[0095] The update rule for the excitatory connection weights between head-facing cells is:

[0096]

[0097] in, is the discharge rate used for learning at time t, k HD =0.1 is the adjustment coefficient; is the discharge rate used for learning at time t-1;

[0098] The calculation method is as follows:

[0099]

[0100] Among them, σ HD is the variance, Indicates the current yaw angle of the drone ψ HD with the head toward the cell's preferred direction The difference between The calculation is as follows:

[0101]

[0102] Self-motion information and external perception information serve as input to head heading cells I m HD :

[0103] I m HD (t) = I m S,ψ (t)+I m E,ψ (t),

[0104] Among them, I m S,ψ is the yaw angle input of the self-motion cue, I m E,ψ The yaw angle input is the external perception information.

[0105] The head orientation cell information fusion model encodes the yaw angle provided by the self-motion clues and the external perception information based on the Gaussian function. The encoding calculation method is as follows:

[0106]

[0107] Among them, σ*,ψ is the variance, ψ * The yaw angle provided by the self-motion clue or the yaw angle provided by the external sensory information; indicates the preferred direction of the head toward the cell; Indicates I m S,ψ (t) or I m E,ψ (t).

[0108] Wherein, the discharge rate Calculate using the hyperbolic tangent function:

[0109]

[0110] The decoding in step 3 is to calculate the yaw angle ψ by using the group vector method to decode the group discharge rate of the cell toward which the head is facing. HD , the process is:

[0111]

[0112]

[0113]

[0114] in, represents the firing rate of the head-direction cell.

[0115] In the encoding process described in step 4, the activity of the three-dimensional periodic grid cell information fusion model is updated as follows:

[0116]

[0117] in, Indicates local excitatory activity, Indicates activity inhibition, ΔP GC represents the activity offset, Represents the input of external perception information; i, j, k represent the three-dimensional cells in the three-dimensional periodic grid cell information fusion model.

[0118] Wherein, the local excitatory activity Excitatory connection weights Build, the build process is as follows:

[0119]

[0120] in, and is the dimension of the three-dimensional periodic grid cell information fusion model;

[0121] Excitatory connection weights The calculation process is as follows:

[0122]

[0123] in, and represents the variance, and Represents three-dimensional periodic grid cell sequences and The position x corresponding to the UAV in the three-dimensional periodic grid cell information fusion model c ,y c and z c the distance between them;

[0124] and The calculation process is as follows:

[0125]

[0126] Wherein, the inhibitory activity The calculation process is as follows:

[0127]

[0128] in, To suppress the weight, Indicates global inhibition; is the dimension of the three-dimensional periodic grid cell information fusion model;

[0129] Activity offset The calculation process is as follows:

[0130]

[0131] Among them, δ x0 , δ y0 , δ z0 is the integer part of the remainder γ, which is the horizontal velocity v obtained by the self-motion clue S , height and speed and yaw angle ψ HD Determined together, the calculation process is as follows:

[0132]

[0133]

[0134]

[0135]

[0136] in, and is the adjustment coefficient; δ xf , δ yf , δ zf is the decimal part of the remainder; a, b are variables in the function f(a,b).

[0137] Among them, the input of the external perception information It is used to correct the accumulated error of the three-dimensional periodic grid cell information fusion model. The three-dimensional periodic grid cell information fusion model encodes the position of the external perception information based on the Gaussian function. The encoding process is as follows:

[0138]

[0139]

[0140] Among them, the mod() function is used to calculate the position x provided by the external perception information E 、y E and z E With parameters and the remainder of and Represents a cyclic sequence based on the positions provided by external sensory information.

[0141] The method for determining the position of the UAV by decoding the motion state of the wave packet in step 5 is:

[0142] Step 501: Calculate the center position of the wave packet (x GC ,y GC ,z GC )as follows:

[0143]

[0144] in, is the dimension of the three-dimensional periodic grid cell information fusion model; is a three-dimensional periodic grid cell sequence;

[0145]

[0146]

[0147]

[0148]

[0149]

[0150]

[0151] k size =2π / n GC ;

[0152] Among them, n GC For the same value or or

[0153] Step 502: According to the change of the position of the wave packet, the wave packet movement speed is obtained as follows:

[0154]

[0155] Step 503: The position (x, y, z) of the UAV is further obtained according to the velocity of the wave packet as follows:

[0156]

[0157] Among them, c x , c y and c z is the adjustment coefficient.

[0158] like Figure 2 and 3 As shown in the figure, the true value represents the real flight trajectory of the UAV. When the three-dimensional periodic grid cell information fusion model has no input of external perception information, the positioning of the UAV will have a large deviation from the true value. Figure 2 When the constructed three-dimensional periodic grid cell information fusion model integrates self-motion clues and external perception information, the positioning accuracy of the UAV is improved, and its positioning result is closer to the true value. Figure 3 shown.

Claims

1. A brain-inspired navigation method that integrates self-motion cues and external sensory information, characterized in that: The following steps are involved: Step 1: The drone's navigation system includes an inertial navigation system and a global navigation satellite system. The inertial navigation system provides egomotion cues, and the global navigation satellite system provides external perception information. The egomotion cues include yaw angle and speed, and the external perception information includes yaw angle and position information. Step 2: A continuous attractor neural network was used to construct a head-facing cell information fusion model to simulate the discharge characteristics of head-facing cells. Specifically, the model encoded the yaw angle of self-motion cues and external sensory information to generate activity changes, which were converted into the group discharge rate of head-facing cells through an activation function. Step 3: Use the group firing rate of the head-facing cells to decode the yaw angle of the drone. The yaw angle of the drone is the fusion result of the yaw angle of the self-motion information and the yaw angle of the external perception information; Step 4: Use a continuous attractor neural network to construct a three-dimensional periodic grid cell information fusion model to simulate the discharge characteristics of three-dimensional periodic grid cells. That is, use this model to encode the yaw angle, speed of self-motion cues, and position of external sensory information decoded in step 3 to produce activity changes; Step 5: Based on the wave packet formed by the activity of the three-dimensional periodic grid cell information fusion model, the position of the UAV is determined by decoding the motion state of the wave packet.

2. The brain-inspired navigation method for fusing self-motion cues and external sensory information according to claim 1 is characterized in that: During the encoding process described in step 2, the activity of the head-facing cell information fusion model Updates as follows: ， in, is the excitatory connection weight between head-facing cells, For local inhibition, represents global suppression, represents the firing rate of the head-facing cell, represents the input of the information fusion model of the head-facing cell, is the weight coefficient, is the time constant; m is the head direction cell at the current moment, and n is the head direction cell at the previous moment; The update rule for the excitatory connection weights between head-facing cells is: ， in, is the discharge rate used for learning at time t, is the adjustment coefficient; is the discharge rate used for learning at time t-1; The calculation method is as follows: ， in, is the variance, Indicates the current yaw angle of the drone with the head toward the cell's preferred direction The difference between The calculation is as follows: ； Self-motion information and external sensory information serve as inputs to head heading cells : ， in, is the yaw angle input of the self-motion cue, The yaw angle input is the external perception information.

3. The brain-inspired navigation method for fusing self-motion cues and external sensory information according to claim 2 is characterized in that: The head orientation cell information fusion model encodes the yaw angle provided by the self-motion clues and the external perception information based on the Gaussian function. The encoding calculation method is as follows: ， in, is the variance, , The yaw angle provided by the self-motion clue or the yaw angle provided by the external sensory information; indicates the preferred direction of the head toward the cell; express or .

4. The brain-inspired navigation method for fusing self-motion cues and external sensory information according to claim 2, characterized in that: The discharge rate Calculate using the hyperbolic tangent function: 。 5. The brain-inspired navigation method for fusing self-motion cues and external sensory information according to claim 2, characterized in that: The decoding described in step 3 is to calculate the yaw angle by using the group vector method to decode the group discharge rate of the cell toward which the head is facing. , the process is: ， ， ； in, represents the firing rate of the head-direction cell.

6. The brain-inspired navigation method for fusing self-motion cues and external sensory information according to claim 1, characterized in that: During the encoding process described in step 4, the activity of the three-dimensional periodic grid cell information fusion model is updated as follows: ， in, Indicates local excitatory activity, Indicates active inhibition, represents the activity offset, Represents the input of external sensory information; Represents three-dimensional cells in the three-dimensional periodic grid cell information fusion model.

7. The brain-inspired navigation method for fusing self-motion cues and external sensory information according to claim 6, characterized in that: The local excitatory activity Excitatory connection weights Build, the build process is as follows: ， in, , and is the dimension of the three-dimensional periodic grid cell information fusion model; Excitatory connection weights The calculation process is as follows: ， in, , and represents the variance, , and Represents three-dimensional periodic grid cell sequences , and The position of the drone corresponding to the three-dimensional periodic grid cell information fusion model , and the distance between them; , and The calculation process is as follows: 。 8. The brain-inspired navigation method for fusing self-motion cues and external sensory information according to claim 6, characterized in that: The activity inhibits The calculation process is as follows: ； in, To suppress the weight, Indicates global inhibition; 、 、 is the dimension of the three-dimensional periodic grid cell information fusion model.

9. The brain-inspired navigation method for fusing self-motion cues and external sensory information according to claim 6, characterized in that: The activity shift It is used to encode the velocity provided by the self-motion clues and the decoded yaw angle. The encoding process is as follows: ； in, remainder The integer part, remainder Horizontal velocity derived from self-motion cues , height and speed and yaw angle Determined together, the calculation process is as follows: ， ， ； ； in, , and is the adjustment coefficient; is the decimal part of the remainder; a, b are functions Variables in The input of the external perception information It is used to correct the accumulated error of the three-dimensional periodic grid cell information fusion model. The three-dimensional periodic grid cell information fusion model encodes the position of the external perception information based on the Gaussian function. The encoding process is as follows: ， ； in, Functions are used to calculate the positions provided by external perception information 、 and With parameters , and the remainder of 、 and Represents a cyclic sequence based on the positions provided by external sensory information.

10. The brain-inspired navigation method for fusing self-motion cues and external sensory information according to claim 6, characterized in that: The method for determining the position of the UAV by decoding the motion state of the wave packet in step 5 is: Step 501: Calculate the center position of the wave packet as follows: ， in, , , is the dimension of the three-dimensional periodic grid cell information fusion model; is a three-dimensional periodic grid cell sequence; ， ， ， ； in, For the same value or or ; Step 502: According to the change of the position of the wave packet, the wave packet movement speed is obtained as follows: ， Step 503: Further obtain the position of the UAV based on the speed of the wave packet as follows: ， in, , and is the adjustment coefficient.

Images