A biomimetic robotic fish swimming control method based on a shallow brain model

Abstract

The application relates to a shallow brain model-based swimming control method of a bionic robotic fish, and relates to the technical field of robotic fish, and comprises the following steps: a shallow brain model is constructed by simulating a biological nervous system and combining a biological neuron information transmission mechanism, so that the motion capability of the bionic robotic fish is improved; the shallow brain model mainly builds hippocampus modules and cerebellum modules, simulates the biological nervous system of a real organism, and makes the bionic robotic fish more like a real fish. By adopting a recursive fractal structure, the efficiency of neural computation is improved, and the robustness of the network is enhanced in a random connection mode.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of bionic fish, and particularly relates to a bionic robotic fish swimming control method based on a shallow brain model. BACKGROUND

[0002] The statements in this section merely provide background information related to the application and do not necessarily constitute prior art.

[0003] The ocean accounts for about 70% of the earth's area and has always contained rich resources, and the seabed world is full of unknown opportunities and challenges. In recent years, bionic robots have gradually become a hot field in scientific and technological research. As one of the representative applications, underwater bionic robotic fish has a wide application prospect in the fields of marine exploration, environmental monitoring, military security and the like. By simulating the behavior of real fish, a new idea is provided for us to deeply understand the motion control of biological bodies.

[0004] Traditional bionic robotic fish usually rely on complex mechanical structures and control algorithms to achieve basic swimming functions. Although great breakthroughs have been made in the structural design and motion ability of bionic robotic fish, how to achieve efficient, accurate and low-energy swimming control is still an important problem that needs to be continuously solved. Many existing control algorithms are often based on preset rules, lack sufficient flexibility and adaptability, and cannot quickly respond to environmental changes like organisms in nature. Secondly, the traditional robotic fish uses multiple motors to drive multiple joints, frequently starts and stops, and the algorithm is not optimized, resulting in high energy consumption. The introduction of the brain-like model opens up a new chapter for the control of bionic robotic fish. By simulating the processing method of the brain neural network, the brain-like model can realize an efficient control algorithm, thereby promoting the development of bionic robotic fish towards intelligence and high efficiency. SUMMARY

[0005] The application starts from the reasons for the robotic fish swimming control scheme:

[0006] In order to solve the above problems, the application provides a bionic robotic fish swimming control method based on a shallow brain model. The application constructs a shallow brain neural network model, simulates the biological nervous system, combines biological neuron information transmission for swimming control, and improves the perception and motion ability of the robotic fish in a dynamic environment. Obstacles are randomly arranged in a simulation environment, and the bionic robotic fish receives sensor data through a perception input layer, which at least includes a distance sensor and an angle sensor, to obtain relative distance and angle information between the robotic fish and the surrounding environmental obstacles. The data is transmitted to a neural computing layer, which is processed by a hippocampus module and a cerebellum module. An output control layer converts the motion control signals generated by the neural computing layer into motion instructions of the robotic fish, to ensure that the robotic fish avoids obstacles and maintains stable swimming.

[0007] In order to achieve the above object, the present application adopts the following technical solutions:

[0008] A bionic robotic fish swimming control method based on a shallow brain model, comprising the following steps:

[0009] The shallow brain model is constructed, including a perception input layer, a neural computation layer and an output control layer, and the neural computation layer mainly includes a hippocampus module and a cerebellum module;

[0010] The perception input layer is used for receiving input data of a sensor, the sensor including at least one distance sensor and one angle sensor, and the input data including relative distance information and angle information of the robotic fish and the surrounding environment;

[0011] The neural computation layer imitates the basic working mechanism of certain specific regions of the brain, is divided according to brain function, especially performs specific tasks such as perception, decision and action control, constructs the hippocampus module and the cerebellum module, perceives the input data and generates output signals and motion control signals;

[0012] The connection mode of neurons of a real organism is simulated, and the neurons in each layer are randomly connected without interfering with each other.

[0013] The output control layer generates motion instructions of the robotic fish according to the control signals generated by the neural computation layer, drives the tail fin and pectoral fin of the robotic fish to perform swimming actions, and realizes swimming control and obstacle avoidance.

[0014] The shallow brain model is used for simulation of bionic robotic fish swimming control, and the motion trajectory of the bionic robotic fish is obtained.

[0015] As an optional implementation, matlab is used for simulation, the hippocampus and the cerebellum model are packaged, and neuron hierarchical connection is built. Obstacles are generated in coordinates and randomly distributed.

[0016] As an optional implementation, the neural computation layer of the shallow brain model adopts a recursive fractal neural network structure, and the calculation efficiency of the neural network is enhanced through hierarchical network design.

[0017] As an optional implementation, the weight between the perception input layer and the neurons of the neural computation layer is , and the weight from the neural computation layer to the output control layer is .

[0018] As an optional implementation, the shallow brain model introduces random connection weights between neurons in each layer, simulates the sparse connection and redundant structure between neurons in the biological nervous system, and improves the network robustness.

[0019] As a further limitation, the shallow brain model control bionic robotic fish swimming control implementation process is:

[0020] Step 101: Distance sensors are installed at different positions of the robotic fish, which measure the relative distance between the robotic fish and the surrounding obstacles in real time through ultrasonic technology.

[0021] As an alternative embodiment, the distance sensor can be selected from ultrasonic sensor, laser sensor or infrared sensor, etc. The distance calculation using ultrasonic technology is represented as:

[0022]

[0023] wherein, is the straight-line distance between the bionic robotic fish and the obstacle, is the round-trip time of the signal, is the propagation speed of the sound wave.

[0024] Step 102: The angle sensor collects the relative angle information between the robotic fish and the obstacle, and updates the angle between the obstacle and the robotic fish in real time according to the motion state of the robotic fish and the environmental conditions.

[0025] Step 103: The data is input into the hippocampus module and the cerebellum module. The hippocampus module processes the distance data and generates the corresponding perception output; the cerebellum module processes the angle data, calculates the motion direction of the robotic fish, and adjusts the motion strategy.

[0026] As an alternative embodiment, the Tansig activation function is used in the neural network for data processing. Through its nonlinear characteristics, it helps the neural network to process complex nonlinear decision-making tasks.

[0027]

[0028] Tansig is a hyperbolic tangent function that maps the output data to , where u is the independent variable of the activation function.

[0029] Step 104: The hippocampus module and the cerebellum module in the neural computing layer independently calculate and integrate their output information to generate the final motion control signal.

[0030] As an alternative embodiment, the specific process of inputting and outputting of the shallow brain model is as follows:

[0031] Step 201: The input of the hippocampus is the Euclidean distance between the robotic fish and the obstacle obtained by the distance sensor:

[0032]

[0033] The difference in the horizontal direction is , and the distance in the vertical direction is The square root of the sum of the square of the difference is the Euclidean distance.

[0034] Step 202: The bionic robotic fish simulates the function of the cerebellum, dynamically adjusts its own motion trajectory and direction according to real-time angle perception data, which can be represented as:

[0035]

[0036] The angle is in radians, indicating the direction from the first point to the second point. and respectively represent the difference in position of the robotic fish and the obstacle on the axis and axis.

[0037] Step 203: Data transmission between the perception input layer and the neural calculation layer:

[0038]

[0039] For a single neuron in the hidden layer , its input is the distance and angle data received by the sensor. The weight is the weight parameter of the th neuron in the input layer and the th neuron in the hidden layer. The bias is , is the activation function, is the output of the hidden layer.

[0040] Step 204: Data transmission between the neural calculation layer and the output control layer is:

[0041]

[0042] where, is the number of output layers, is the weight between the output layer neuron and the hidden layer neuron , is the bias term of the output layer neuron . is the final output result.

[0043] The application builds a shallow brain model and controls the swimming of the bionic robotic fish based on the shallow brain model, so that the bionic robotic fish can autonomously avoid obstacles. The shallow brain model simplifies and simulates the lower-level neural activity in the biological nervous system, providing an efficient control and calculation framework. Generally, neurons are divided into three different levels, including a perception input layer, a neural calculation layer and an output control layer, and each level of neuron is responsible for processing different types of information and interacting with neurons of other levels. In addition, the connection between neurons is randomized, that is, the connection relationship between neurons is not set in advance, but is randomly generated based on a certain probability, breaking the limitation of full connection in the traditional neural network, so that the network can dynamically adjust the connection in different environments, thereby increasing the adaptability and robustness of the network. The sparse connection between neurons reduces the number of connections between neurons, reduces the computational complexity and memory requirements. The neural network adopts a recursive fractal structure, and each level of neuron is connected to the previous layer through recursion, which has a fractal-like feature in structure and can quickly calculate and process through the structure.

[0044] The application has the following beneficial effects:

[0045] The application provides a bionic robotic fish swimming control method based on a shallow brain model, which simulates the working principle of the biological nervous system to realize the perception and adaptive obstacle avoidance of the bionic robotic fish. Through the recursive fractal neural network structure, the shallow brain model can efficiently process data of different sensors and quickly generate motion control signals, so that the robotic fish can flexibly adjust the motion direction to avoid obstacles in a complex environment. Secondly, the shallow brain model has good generalization ability and strong adaptability, and can maintain good motion control performance in a dynamically changing environment. In addition, the randomized connection of neurons enhances the fault tolerance of the network, so that the system can avoid falling into a local optimal solution, thereby improving the overall robustness. BRIEF DESCRIPTION OF DRAWINGS

[0046] The drawings accompanying the specification of the application form a part thereof and serve to further understand the application, the illustrative embodiments of the application and the description thereof serve to explain the application without constituting an improper limitation thereof.

[0047] Figure 1 The structural block diagram of the shallow brain model of at least one embodiment of the application;

[0048] Figure 2 The information processing schematic diagram of the shallow brain model of at least one embodiment of the application;

[0049] Figure 3 The structural schematic diagram of the shallow brain model for controlling the bionic robotic fish of at least one embodiment of the application; DETAILED DESCRIPTION

[0050] The present application will be further described with reference to the drawings and examples.

[0051] It should be noted that the following detailed description is illustrative only, and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0052] It is also important to note that the terms "including", "containing", and / or "comprising", when used in this specification, specify the presence of stated features, steps, operations, devices, components, and / or combinations thereof, but do not preclude the presence or addition of one or more other features, steps, operations, devices, components, and / or groups thereof.

[0053] A bionic robotic fish swimming control method based on a shallow brain model, comprising the following steps:

[0054] S1: Constructing a shallow brain model, including a perception input layer, a neural computation layer, and an output control layer, the neural computation layer mainly including a hippocampus module and a cerebellum module;

[0055] S2: The perception input layer is used to receive input data of sensors, the sensors including at least one distance sensor and one angle sensor, and the input data including relative distance information and angle information of the robotic fish and the surrounding environment;

[0056] S3: The neural computation layer simulates the basic working mechanism of certain specific regions of the brain, and is divided into brain function zones, especially for performing specific tasks such as perception, decision-making, and action control, to construct the hippocampus module and the cerebellum module;

[0057] S4: Simulating the connection mode of neurons of real organisms, and randomly connecting between neurons of each layer without interference between neurons within the layer.

[0058] S5: The output control layer generates a motion instruction of the robotic fish according to the control signal generated by the neural computation layer, to drive the tail fin and the pectoral fin of the robotic fish to perform swimming actions, and to realize swimming control and obstacle avoidance.

[0059] S6: The shallow brain model is used for simulation of bionic robotic fish swimming control, to obtain a motion trajectory of the bionic robotic fish.

[0060] Specifically, the shallow brain neural network model in step S1 includes a perception input layer, a neural computation layer, and an output control layer, and the model is a recursive classification structure, and each level of network or module has certain self-similarity in structure.

[0061] Specifically, the sensor and the bionic robotic fish swimming control implementation process in step S2 are as follows:

[0062] Step 101: The distance sensor is installed at different positions of the robotic fish, and the relative distance between the robotic fish and the surrounding obstacles is measured in real time through ultrasonic technology.

[0063] As an alternative embodiment, the distance sensor can be an ultrasonic sensor, a laser sensor, or an infrared sensor, etc., and the distance calculation using ultrasonic technology is represented as:

[0064]

[0065] Wherein, is the straight-line distance between the bionic robotic fish and the obstacle, is the round-trip time of the signal, is the propagation speed of the sound wave.

[0066] Step 102: The angle sensor collects the relative angle information between the robotic fish and the obstacle, and updates the angle between the obstacle and the robotic fish in real time according to the motion state of the robotic fish and the environmental conditions.

[0067] Step 103: The data is input into the hippocampus module and the cerebellum module, the hippocampus module processes the distance data and generates the corresponding perception output, and the cerebellum module processes the angle data and calculates the motion direction adjustment strategy of the robotic fish.

[0068] As an alternative embodiment, the Tansig activation function is used in the neural network for data processing. Through its nonlinear characteristics, it helps the neural network to process complex nonlinear decision-making tasks.

[0069]

[0070] Tansig is a hyperbolic tangent function that maps the output data to , where u is the independent variable of the activation function.

[0071] Step 104: The hippocampus module and the cerebellum module in the neural computing layer are independently calculated, and the final motion control signal is generated after integrating the output information of each module.

[0072] Specifically, the basic working mechanism of some specific regions in step S3 is simulated, the hippocampus module and the cerebellum module are modeled according to brain function partitioning, and other modules such as the basal ganglia and the thalamus can be added to make the shallow brain model more perfect.

[0073] Specifically, the shallow brain neural network model in step S4 is connected between each layer of neurons through a certain probability of random connection, and the random connection is sparse, which reduces redundant calculation, avoids overfitting, and improves efficiency.

[0074] Specifically, the specific flow of the shallow brain model input to output in step S5 is as follows:

[0075] Step 201: The input of the hippocampus is the Euclidean distance between the robotic fish and the obstacle obtained by the distance sensor:

[0076]

[0077] The difference in the horizontal direction is , and the distance in the vertical direction is . The Euclidean distance is obtained by calculating the square sum of the difference and taking the square root.

[0078] Step 202: The bionic robotic fish dynamically adjusts its motion trajectory and direction according to real-time angle perception data by simulating the function of the cerebellum, which can be represented as:

[0079]

[0080] The angle is in radians, representing the direction from the first point to the second point. and represent the difference in position of the robotic fish and the obstacle on the axis and axis, respectively.

[0081] Step 203: Data transmission between the perception input layer and the neural calculation layer:

[0082]

[0083] For a single neuron in the hidden layer , its input is the distance and angle data received by the sensor. The weight is the weight parameter of the th neuron in the input layer and the th neuron in the hidden layer. The bias is , is the activation function, is the output of the hidden layer.

[0084] Step 204: Data transmission between the neural calculation layer and the output control layer:

[0085]

[0086] where, is the number of output layers, is the weight between the output layer neuron and the hidden layer neuron . is the bias term of the output layer neuron . is the final output result.

Claims

1. A shallow brain model-based biomimetic robotic fish swimming control method, characterized by: Comprising the following steps: The shallow brain model is constructed to simulate the basic working mechanism of certain specific regions of the brain, and is modeled according to brain function division to perform specific tasks including perception, decision-making and action control; The shallow brain neural network model includes a perception input layer, a neural computing layer and an output control layer; The shallow brain model is constructed to include a perception input layer, a neural computing layer and an output control layer, and the neural computing layer includes a hippocampus module and a cerebellum module; The perception input layer is used to receive input data of sensors, which include at least one distance sensor and one angle sensor, and the input data includes relative distance information and angle information of the robotic fish and the surrounding environment; The neural computing layer simulates the basic working mechanism of certain specific regions of the brain, and is divided according to brain function to construct a hippocampus module and a cerebellum module, which perceive input data and generate output signals and motion control signals; The hippocampus module and the cerebellum module in the neural computing layer are implemented through a recursive fractal neural network structure, which enhances the adaptability and computational efficiency of the neural network through recursive hierarchy and self-similarity; The hippocampus module uses a weighted and activated function of a feedforward neural network to process input data from the distance sensor to generate a distance perception signal, which outputs the Euclidean distance between the robotic fish and the obstacle; The cerebellum module processes input data from the angle sensor through a feedforward neural network and generates a motion direction adjustment strategy for the robotic fish according to the calculated relative angle signal; The connection mode of neurons of a real organism is simulated, and the neurons in each layer are randomly connected without interfering with each other; The output control layer generates motion instructions for the robotic fish according to the control signals generated by the neural computing layer to drive the tail fin and pectoral fin of the robotic fish to perform swimming actions, thereby achieving swimming control and obstacle avoidance; The shallow brain model is used to simulate the swimming control of the biomimetic robotic fish to obtain the motion trajectory of the biomimetic robotic fish.

2. The bionic robotic fish swimming control method based on a shallow brain model according to claim 1, characterized in that: The implementation process of the shallow brain model for controlling the swimming control of the biomimetic robotic fish is as follows: Step 101: Distance sensors are installed at different positions of the robotic fish, and the relative distance between the robotic fish and the surrounding obstacles is measured in real time through ultrasonic technology; Step 102: The angle sensor collects the relative angle information between the robotic fish and the obstacle, and updates the angle between the obstacle and the robotic fish in real time according to the motion state of the robotic fish and the environmental conditions; Step 103: The data is input into the hippocampus module and the cerebellum module, the hippocampus module processes the distance data and generates corresponding perception output; The cerebellum module processes the angle data, calculates the motion direction of the robotic fish, and adjusts the motion strategy; Step 104: The hippocampus module and the cerebellum module in the neural computing layer independently calculate and integrate the output information to generate the final motion control signal.

3. The bionic robotic fish swimming control method based on a shallow brain model according to claim 1, characterized in that: Random connection weights are introduced between the neurons in each layer of the shallow brain model to simulate the sparse connection and redundant structure between neurons in the biological nervous system, improve the network robustness, and enhance the adaptive ability of the network.

4. The bionic robotic fish swimming control method based on a shallow brain model according to claim 1, characterized in that: The implementation process of the shallow brain model is as follows: Step 201: The input of the hippocampus module is the Euclidean distance between the robotic fish and the obstacle obtained by the distance sensor: ； The difference in the horizontal direction is , and the distance in the vertical direction is ; Step 202: The bionic robotic fish simulates the function of the cerebellum, dynamically adjusts its own motion trajectory and direction according to real-time angle perception data, which is represented as: ； angle in radians, representing the direction from the first point to the second point; and respectively represent the position difference of the robotic fish and the obstacle on the axis and axis; Step 203: Data transmission between the perception input layer and the neural computing layer: ； For a single neuron In the hidden layer whose inputs are the distance and angle data received by the sensors; the weights are the weight parameters of the input layer's first neuron and the hidden layer's first neuron; bias is , is an activation function, is the output of the hidden layer; Step 204: Data transmission between the neural computing layer and the output control layer is: ； wherein, is the number of output layers, is the weight between the output layer neuron and the hidden layer neuron , is the bias term of the output layer neuron ; is the final output result.

Images