An artificial intelligence-based bionic robot control logic acquisition method

Abstract

This invention belongs to the interdisciplinary field of artificial intelligence methods and biomimetic technology, and relates to an AI-based method for acquiring the control logic of biomimetic robots. From the perspective of biomimetic robot control logic, its advantage lies in the fact that existing biomimetic robots merely mimic biological movement postures, making it difficult to achieve the energy-efficient, high-efficiency, and low-noise movements of living organisms, thus failing to meet current task requirements. Therefore, an AI-based method for acquiring the control logic of biomimetic robots is designed. This method uses artificial intelligence to obtain the control logic of biomimetic robots from real biological movement processes, aiming to improve the overall performance of biomimetic robots. Biomimetic robots that acquire control logic using this method can move more energy-efficiently and with lower noise, better meeting the needs of various current tasks and enhancing their practical value.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of artificial intelligence methods and bionic technology, and relates to a method for obtaining control logic of bionic robots based on artificial intelligence. Background Technology

[0002] With the increasing demands of land and underwater operations, especially ecological surveys requiring stealth, various biomimetic robots have emerged. However, existing biomimetic robots merely mimic the postures of biological movements, failing to achieve the same high efficiency, energy saving, and high stealth (low noise) as living organisms. With the development of environmental information measurement technology, sensor technology, and artificial intelligence technology, more detailed control of robot movements that closely resembles real biological motion has become possible. Therefore, designing an AI-based method for acquiring control logic for biomimetic robots to comprehensively improve their motion performance is essential. Summary of the Invention

[0003] The purpose of this invention is to overcome the problems of high energy consumption, low efficiency and high noise caused by the difficulty of the propulsion control logic of existing bionic robots in mimicking real organisms. This invention proposes an artificial intelligence-based method for obtaining the control logic of bionic robots, which comprehensively improves the performance of bionic robots.

[0004] The objective of this invention is achieved through the following technical solution:

[0005] A method for acquiring control logic of a biomimetic robot based on artificial intelligence, comprising the following steps:

[0006] Step 1: Use environmental information measurement technology and sensors to acquire data on the movement of the organism to be imitated and the surrounding environment, and acquire data on different movement states, especially the adjustments made by the organism in response to different environmental conditions and changes in direction.

[0007] A PIV experimental setup was constructed to obtain information on the body shape changes and surrounding flow field of organisms undergoing linear acceleration swimming in still water using the PIV method.

[0008] Step 2: Analyze the mechanical joint structure of the control system required to mimic biological body posture. Based on the proposed control system, process the acquired movement posture data, velocity field and derived vector diagram. Use environmental information and movement state as input information for the neural network, and the corresponding signal of the control system as the label.

[0009] Step 3: Train the neural network, judge based on the loss function value during the training process, and obtain the network model and parameters with the best performance, which will be used as the control logic of the bionic robot.

[0010] Step 4: Use numerical simulation software to verify whether the bionic robot control logic obtained by the artificial intelligence method can provide effective feedback based on the set movement state and surrounding environment information.

[0011] Furthermore, the specific steps of step 1 are as follows:

[0012] Step 1.1: The PIV experimental equipment was set up. To ensure that the linear acceleration swimming information of the organisms could be obtained more easily, it was confined in a transparent water tank. Two days before the formal experiment, the organisms were placed in the water tank to adapt to the water tank environment.

[0013] Step 1.2: Install two continuous lasers located on both sides of the pool to eliminate the shadow areas created by a single laser;

[0014] Step 1.3: The image acquisition system is a high-speed camera, and polyamide particles are evenly distributed in the water tank as tracer particles;

[0015] Step 1.4: The laser system and the image acquisition system are fixed on two multi-degree-of-freedom optical platforms respectively. The high-speed camera's shooting angle is perpendicular to the laser plane. The experiment captures the autonomous propulsion movements of the organism in the data acquisition area without applying any external stimuli.

[0016] Step 1.5: Consider the need to obtain biological swimming data in the most natural state and the memory of the high-speed camera. After the organism swims into the data acquisition area autonomously and is at the height of the laser plane, start shooting, acquire raw image data and save it.

[0017] Step 1.6: The acquired raw image data is preprocessed, the preprocessed image is calibrated, and then adaptive PIV calculation is used to obtain the final body shape diagram, velocity field and derived vector diagram of the organism; finally, the body shape changes of the organism swimming in still water with linear acceleration and the surrounding flow field information are obtained.

[0018] Furthermore, step 2 is specifically as follows:

[0019] Step 2.1: Analyze the changes in the body posture of the organism to obtain a suitable mechanical joint structure for the overall control system of the biomimetic robot, ensuring that the degree of freedom of the control system can reproduce the swinging process of the real organism to the greatest extent. This requires N transmission nodes and corresponding inter-joint transmission rods and shell support transmission rods to reproduce the swinging process of the real organism.

[0020] Step 2.2: Process the body posture data during movement according to the proposed overall control system structure of the bionic robot. Organize the states that each transmission node should be in into a signal matrix, which serves as a label during the neural network training process. In the signal matrix, A1-A5 are responsible for controlling the posture of the bionic robot. A1 determines the orientation of the head, A5 determines the swing direction of the tail, and A2-A4 determine the swing angle of the inter-joint transmission rods of the N central transmission nodes along the longitudinal direction of the bionic robot. In the signal matrix, R2-R4 and L2-L4 determine the swing angles of the right and left transmission rods of the N central transmission nodes, respectively. In application, the matrix can be set to any form as needed to ensure reasonable control of the bionic robot.

[0021] Step 2.3: Organize the environmental information around the organism and use it as input information for the neural network to enable the bionic robot to measure and acquire the corresponding position data; based on the location of the flow sensor of the designed bionic robot, read r1-r at different times from the velocity field and derived vector field. 10 and l1-l 10 The corresponding information forms a dataset of flowing information matrix input to the neural network.

[0022] Furthermore, step 3 is as follows:

[0023] Step 3.1: Train the neural network using GPU acceleration so that it can determine the signal matrix at the next moment by inputting the flow information and signal matrix at several consecutive time steps. During training, the change of the loss function value should be obtained.

[0024] Step 3.2: Based on the changes in the loss function value, determine the optimal neural network model and parameters, which will serve as the control logic for the biomimetic robot.

[0025] Furthermore, step 4 is as follows:

[0026] Step 4.1: Model the bionic robot using the modeling software CATIA and import it into the numerical simulation software Fluent for simulation. Determine whether the acquired control logic can effectively respond appropriately to the flow field information, and parameters such as the efficiency of propulsion using this method.

[0027] Step 4.2: Make corrections based on the numerical simulation results to obtain the optimal network model and parameters.

[0028] The beneficial effects of this invention are as follows:

[0029] This invention addresses the issue from the perspective of biomimetic robot control logic. Its advantage lies in recognizing that existing biomimetic robots merely mimic biological postures, failing to achieve the energy-efficient, high-efficiency, and low-noise movements of living organisms, thus not meeting current task requirements. Therefore, this invention designs an artificial intelligence-based method for acquiring biomimetic robot control logic. By utilizing artificial intelligence to extract the control logic from real biological movements, the invention aims to improve the overall performance of the biomimetic robot. Biomimetic robots using this method can move more energy-efficiently and with lower noise, better meeting the demands of various current tasks and enhancing their practical value. Attached Figure Description

[0030] Figure 1 Schematic diagram of the PIV test setup.

[0031] Figure 2 PIV image processing procedure.

[0032] Figure 3 A schematic diagram of the proposed zebrafish overall control system structure.

[0033] Figure 4 Partial schematic diagram of the transmission node.

[0034] Figure 5 signal matrix

[0035] Figure 6 Flow field information matrix

[0036] Figure 7 Neural network input and output diagram

[0037] Figure 8 A schematic diagram of the control logic acquisition and practical application process. Detailed Implementation

[0038] The present invention will now be further described with reference to the accompanying drawings.

[0039] This invention discloses a method for acquiring control logic for biomimetic robots based on artificial intelligence. The method includes acquiring data on the movement of the simulated organism and its surrounding environment, analyzing the requirements of the biomimetic robot control system and organizing the training dataset, determining the neural network model and parameters, and verifying the data using numerical simulation software. Taking the linear acceleration of a zebrafish in still water as an example, environmental information is acquired using particle image velocimetry (PIV). The specific steps are as follows:

[0040] Step 1: Set up a PIV experimental setup and use the PIV method to obtain information on the body shape changes of zebrafish 2 swimming linearly with acceleration in still water and the surrounding flow field.

[0041] Step 1.1: The PIV test equipment was set up. To ensure that the linear acceleration swimming information of zebrafish 2 was more easily obtained, it was confined in a transparent water tank 1 with dimensions of length × width × height = 400 * 200 * 400 mm. The water was only filled to 50 mm, and the water temperature was controlled at 20.6 ± 0.5℃, which is more active for zebrafish 2. Two days before the formal test, zebrafish 2 were placed in water tank 1 to adapt to the water tank environment.

[0042] Step 1.2: Install two 10W Nd:YAG continuous lasers 3 with a wavelength of 532 nm on both sides of the pool to eliminate the shadow area generated by a single laser. The thickness of the light source is 1 mm and the distance between the light source and the bottom of the pool is 25 mm.

[0043] Step 1.3: The image acquisition system was a Photron-FASTCAM_Mini_UX100 CCD high-speed camera 4 with a resolution of 1024 pixels × 1024 pixels and a measurement area of ​​96.7 mm × 96.7 mm. The shooting frame rate used in the experiment was 2000 FPS. Polyamide particles with an average diameter of 10 μm were evenly scattered in the water tank as tracer particles.

[0044] Step 1.4: The laser system and image acquisition system are fixed on two multi-degree-of-freedom optical platforms respectively. The high-speed camera 4 has a shooting angle perpendicular to the laser plane. The experiment captures the autonomous propulsion movements of zebrafish 2 within the data acquisition area 5, without any external stimulation. The specific PIV experimental setup is as follows: Figure 1 As shown.

[0045] Step 1.5: Considering the need to obtain the most natural swimming data of zebrafish 2 and the limited memory of high-speed camera 4, we need to patiently wait for zebrafish 2 to swim into the data acquisition area 5 on its own and be at the height of the laser plane before starting to take pictures, acquire the original image data 6 and save it.

[0046] Step 1.6: The acquired raw image data 6 is preprocessed, and the preprocessed image 7 is calibrated 8. Then, adaptive PIV calculation 9 is used to obtain the final zebrafish body posture diagram, velocity field, and derived vector diagram. The PIV image processing process is as follows: Figure 2 As shown, the final results obtained include the body shape changes of zebrafish 2 during linear acceleration in still water and information about the surrounding flow field.

[0047] Step 2: Analyze the mechanical joint structure of the control system required for the zebrafish 2 body posture. Based on the proposed control system, process the acquired movement posture data, velocity field and derived vector diagram. Use environmental information as input information for the neural network and the corresponding signal of the control system as a label.

[0048] Step 2.1: Analyze the body posture changes of zebrafish 2 to obtain a suitable mechanical joint structure for the overall control system of the biomimetic zebrafish body 11. This ensures that the degrees of freedom of the control system can reproduce the swinging process of a real zebrafish to the greatest extent. If five transmission nodes and corresponding transmission rods are required to reproduce the swinging process of a real zebrafish, then the corresponding structure is as follows: Figure 3 As shown. The transmission node is as follows. Figure 4 As shown.

[0049] Step 2.2: Process the body posture data during movement according to the overall control system structure of the proposed biomimetic zebrafish main body 11. Organize the states that each transmission node 12 should be in into a signal matrix form, which will serve as labels in the neural network training process. The signal matrix is ​​as follows: Figure 5 As shown in the diagram, signal matrices A1-A5 control the posture of the biomimetic zebrafish body 11. A1 determines the orientation of the mechanical fish's head, A5 determines the swing direction of the mechanical fish's tail, and A2-A4 determine the swing angles of the transmission rods 13 between the three central transmission nodes along the longitudinal joints of the mechanical fish body. Signal matrices R2-R4 and L2-L4 respectively determine the swing angles of the outer shell support transmission rods 14 on the right and left sides of the three central transmission nodes. In practical applications, the matrix can be set to any form as needed, ensuring reasonable control of the biomimetic robot.

[0050] Step 2.3: Organize the environmental information around the zebrafish 2 as input information for the neural network. Emphasis should be placed on the corresponding positional data that the bionic zebrafish body 11 can easily measure and acquire in practical applications. If the designed bionic zebrafish body 11 is bilaterally symmetrical, a total of twenty flow sensors will be used to acquire actual information. The flow information matrix is ​​as follows: Figure 6 As shown. Based on the location of the flow sensor in the designed biomimetic zebrafish body 11, different times r1-r are read from the velocity field and derived vector field. 10 and l1-l 10 The corresponding information forms a dataset of flowing information matrix input to the neural network.

[0051] Step 3: Train the neural network, judge based on the loss function value during the training process, and obtain the network model and parameters with the best performance, which will be used as the control logic of the bionic robot.

[0052] Step 3.1: Using GPU acceleration, train the neural network so that it can determine the signal matrix at the next time step by inputting the flow information and signal matrix at several consecutive time steps. A schematic diagram of the neural network input and output is shown below. Figure 7 As shown, during training, it is necessary to obtain the changes in the loss function value during the training process.

[0053] Step 3.2: Determine the optimal neural network model and parameters based on the changes in the loss function value, and use them as the control logic for the biomimetic zebrafish main body 11.

[0054] Step 4: Use numerical simulation software to verify whether the bionic robot control logic obtained by the artificial intelligence method can provide effective feedback based on the set movement state and surrounding environment information.

[0055] Step 4.1: Model the biomimetic zebrafish body 11 using the modeling software CATIA, and import it into the numerical simulation software Fluent for simulation to determine whether the acquired control logic can effectively respond to the flow field information and the efficiency of propulsion using this method.

[0056] Step 4.2: Based on the numerical simulation results, make corrections to obtain the optimal network model and parameters. This method is relevant to the practical application of biomimetic fish, such as... Figure 8 As shown.

[0057] Although embodiments of the present invention have been shown and described above, they are merely exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments without departing from the principles and spirit of the present invention. In particular, the prior art is relatively complex in acquiring data on the wagging of fins during swimming. The subsequent acquisition of more swimming states and fin-related data is destined to enable biomimetic robot technology to achieve a leap forward.

[0058] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for acquiring control logic of a biomimetic robot based on artificial intelligence, characterized in that: The steps are as follows: Step 1: Use environmental information measurement technology and sensors to acquire data on the movement of the simulated organism and the surrounding environment, and obtain data under different movement states; A PIV experimental setup was constructed to obtain information on the body shape changes and surrounding flow field of organisms undergoing linear acceleration swimming in still water using the PIV method. Step 2: Analyze the mechanical joint structure of the control system required to mimic biological body posture. Based on the proposed control system, process the acquired movement posture data, velocity field and derived vector diagram. Use environmental information and movement state as input information for the neural network, and the corresponding signal of the control system as the label. Step 2.1: Analyze the changes in the body posture of the organism to obtain a suitable mechanical joint structure for the overall control system of the biomimetic robot, ensuring that the degree of freedom of the control system can reproduce the swinging process of the real organism to the greatest extent. This requires N transmission nodes and corresponding inter-joint transmission rods and shell support transmission rods to reproduce the swinging process of the real organism. Step 2.2: Process the body posture data during movement according to the proposed overall control system structure of the bionic robot. Organize the states that each transmission node should be in into a signal matrix, which serves as a label during the neural network training process. In the signal matrix, A1-A5 are responsible for controlling the posture of the bionic robot. A1 determines the orientation of the head, A5 determines the swing direction of the tail, and A2-A4 determine the swing angle of the inter-joint transmission rods of the N central transmission nodes along the longitudinal direction of the bionic robot. In the signal matrix, R2-R4 and L2-L4 determine the swing angles of the right and left transmission rods of the N central transmission nodes, respectively. In application, the matrix can be set to any form as needed to ensure reasonable control of the bionic robot. Step 2.3: Organize the environmental information around the organism and use it as input information for the neural network to enable the bionic robot to measure and acquire the corresponding position data; based on the location of the flow sensor of the designed bionic robot, read r1-r at different times from the velocity field and derived vector field. 10 and l1-l 10 The corresponding information forms a dataset of flowing information matrix input to the neural network; Step 3: Train the neural network, judge based on the loss function value during the training process, and obtain the network model and parameters with the best performance, which will be used as the control logic of the bionic robot. Step 3.1: Train the neural network using GPU acceleration so that it can determine the signal matrix at the next moment by inputting the flow information and signal matrix at several consecutive time steps. During training, the change of the loss function value should be obtained. Step 3.2: Based on the changes in the loss function value, determine the optimal neural network model and parameters, which will serve as the control logic for the biomimetic robot; Step 4: Use numerical simulation software to verify whether the bionic robot control logic obtained by the artificial intelligence method can provide effective feedback based on the set movement state and surrounding environment information.

2. The method for acquiring control logic of a bionic robot based on artificial intelligence according to claim 1, characterized in that: The specific steps of step 1 are as follows: Step 1.1: The PIV experimental equipment was set up. To ensure that the linear acceleration swimming information of the organisms could be obtained more easily, it was confined in a transparent water tank. Two days before the formal experiment, the organisms were placed in the water tank to adapt to the water tank environment. Step 1.2: Install two continuous lasers located on both sides of the pool to eliminate the shadow areas created by a single laser; Step 1.3: The image acquisition system is a high-speed camera, and polyamide particles are evenly distributed in the water tank as tracer particles; Step 1.4: The laser system and the image acquisition system are fixed on two multi-degree-of-freedom optical platforms respectively. The high-speed camera's shooting angle is perpendicular to the laser plane. The experiment captures the autonomous propulsion movements of the organism in the data acquisition area without applying any external stimuli. Step 1.5: Consider the need to obtain biological swimming data in the most natural state and the memory of the high-speed camera. After the organism swims into the data acquisition area autonomously and is at the height of the laser plane, start shooting, acquire raw image data and save it. Step 1.6: The acquired raw image data is preprocessed, the preprocessed image is calibrated, and then adaptive PIV calculation is used to obtain the final body shape diagram, velocity field and derived vector diagram of the organism; finally, the body shape changes of the organism swimming in still water with linear acceleration and the surrounding flow field information are obtained.

3. The method for acquiring control logic of a bionic robot based on artificial intelligence according to claim 1, characterized in that: Step 4 is as follows: Step 4.1: Model the bionic robot using the modeling software CATIA and import it into the numerical simulation software Fluent for simulation. Determine whether the acquired control logic can effectively respond appropriately to the flow field information, and the parameters when using this method for propulsion. Step 4.2: Make corrections based on the numerical simulation results to obtain the optimal network model and parameters.

Images