An underwater unmanned platform high-reliability intelligent optical positioning method

Abstract

The present application relates to a kind of underwater unmanned platform high reliability intelligent optical positioning method, belong to the technical field of autonomous underwater vehicle autonomous docking and recovery.The present application establishes AUV three-dimensional position angle-coordinate data set and AUV three-dimensional position solution network model, designs AUV three-dimensional position solution geometry loss function, realizes fast and accurate AUV three-dimensional position solution.The angle error of model position solution is 0.00013 ° , meet the requirement of positioning accuracy in underwater unmanned platform guide and dock.The hardware relied on by this method is multi-quadrant sensor, with the advantages of small volume, small amount of detection data, equipment installation is simple, etc., this method also has the advantages of small hardware operation amount, position solution is rapid during deployment.

Description

Technical Field

[0001] This invention relates to a highly reliable intelligent optical positioning method for underwater unmanned platforms, belonging to the field of autonomous underwater vehicle docking and recovery technology. Background Technology

[0002] Autonomous Underwater Vehicles (AUVs) are crucial tools for marine exploration and resource development, and essential components of underwater space stations and seabed observation networks. Currently, the technological development level of AUVs cannot meet the demands of large-scale, widespread application; autonomous, collaborative, and navigation planning technologies require further development and improvement. Among these, autonomous technology is key to the safe and accurate execution of long-term operational tasks by AUVs. Highly reliable intelligent agent cooperative localization and autonomous docking methods for multi-dock, multi-body scenarios place new demands on the accuracy and speed of guidance and pose calculation. Furthermore, the interaction mechanism between multi-dock verification and identification and multi-AUV collaboration is a significant characteristic of multi-dock, multi-body scenarios, and is of practical importance for improving the docking speed and success rate of AUV systems. Researching intelligent agent cooperative localization and autonomous recovery methods for multi-dock, multi-body scenarios is of great significance for establishing an underwater security strategy based on underwater space stations and seabed observation networks.

[0003] From the perspective of AUV guidance methods, they can be mainly divided into four types: acoustic, electromagnetic, optical, and visual. Acoustic guidance has a long range but lower accuracy, low data update rate, and is susceptible to noise interference; electromagnetic and visual guidance typically have a range of... The accuracy is at the centimeter level. Visual guidance is susceptible to changes in background light, seawater refraction and reflection, and turbidity; electromagnetic guidance is susceptible to electromagnetic noise from the AUV itself and interference from external magnetic fields. In practical applications, the effective distance of visual guidance is often... Its effectiveness is limited. In terms of hardware, the monocular or binocular cameras used in visual guidance are large, expensive, and require precise installation. The sensor data generated by visual guidance is a video stream, producing a data volume of [data volume per second]. The sheer volume of data presents a significant challenge for storage and retrieval. Visual guidance pose calculation algorithms are computationally intensive and time-consuming, placing high demands on the computing platform. At higher AUV speeds, the calculation speed often falls short of mission requirements. While optical guidance is also susceptible to variations in background light, seawater refraction and reflection, and turbidity, its smaller sensor size and smaller data volume allow for faster, higher-precision positioning through innovative design of the guidance light source and positioning algorithm. Summary of the Invention

[0004] Technical problems to be solved

[0005] In order to overcome the problems of existing technologies in AUV autonomous technology, such as guidance methods, hardware costs, and pose calculation rates, and to improve the autonomous usability of AUVs, this invention proposes a highly reliable intelligent optical positioning method for underwater unmanned platforms.

[0006] Technical solution

[0007] The objective of this invention is achieved through the following technical solutions and apparatus: establishing an AUV 3D pose angle-coordinate dataset, training an AUV 3D pose calculation network model, and deploying hardware to calculate the AUV's 3D pose. During the AUV pose calculation process, voltage data from multi-quadrant sensors serves as input, and the AUV's real-time relative position is the output. The docking station's beacon system consists of three light sources arranged in an equilateral triangle at the docking station's return port. A large-view area array sensor is the main sensor for optical guidance, positioned at the AUV's head, responsible for receiving the light intensity signal from the docking station's beacon system. The fast 3D pose calculation algorithm uses the light intensity signal measured by the large-view area array sensor as input and is specifically optimized for it. Based on the pre-defined geometric relationships of the beacon system's light sources, it first calculates the angles between the AUV and each light source, and then uses the algorithm to calculate the distance between the AUV and the docking station's return port and the AUV's relative position coordinates. The core performance indicators of the fast 3D pose calculation algorithm are accuracy and speed. By building a small-scale AUV coordinate calculation model and establishing a large-scale AUV underwater positioning angle coordinate dataset through simulation, the parameters of the AUV coordinate calculation model are optimized in an optimized manner, and a multi-dimensional geometric loss relationship is designed to achieve fast pose calculation based on a large-view area array sensor.

[0008] 1. Establish an AUV 3D pose angle-coordinate dataset

[0009] 1) Establish the beacon system coordinate system; the beacon system is located on the recovery dock, and the three guiding light sources are arranged in an equilateral triangle with a side length of [missing information]. The divergence angle of the guiding light source is A rectangular coordinate system is established using the incenter of an equilateral triangle as the origin of the docking station coordinate system; the coordinates of the three guiding light sources are respectively... , , ;

[0010] 2) Calculate AUV angle data based on AUV coordinate data: Assume the AUV is located at coordinate... At this point, the vector , , Solving for , , ;

[0011] 3) Establish an AUV 3D pose angle-coordinate dataset: constrain AUV coordinates The coordinate range, denoted as For the farthest distance of guidance, then , , Within the coordinate range of the AUV, uniform sampling is used to obtain the value of a coordinate. The corresponding angle data is calculated from step 2). At this point, a set of three-dimensional pose angle-coordinate data of the AUV is obtained. ;

[0012] Then: Train the AUV 3D pose calculation network model, including the following sub-steps:

[0013] 1) Establish an AUV 3D pose calculation network model; the AUV 3D pose calculation network model is a multilayer perceptron network or a Transformer regression network model;

[0014] 2) Train the AUV 3D pose solving network. The input of the AUV 3D pose solving network is the angle information in the AUV 3D pose angle-coordinate dataset, and the output is the solved coordinate position. The loss function is the mean square error of the geometric relationship between the solved coordinate position and the actual coordinate position. Use deep learning algorithms to optimize the network parameters and save the trained model parameters.

[0015] Finally: Hardware deployment for calculating the AUV's 3D pose includes the following sub-steps:

[0016] 1) The multi-quadrant detector calculates the guide light angle of the beacon system; the voltage generated by the guide light source in the beacon system in the multi-quadrant detector of the AUV head is... The horizontal deflection angle is determined by the following formula. with vertical deflection angle Angle:

[0017] ;

[0018] in, , , , For multi-quadrant sensors, the calibration value is a constant;

[0019] 2) Calculate the angle between any two guide lights and the center of the multi-quadrant detector; the angle is calculated using the following formula:

[0020]

[0021] , , , , , The horizontal and vertical deflection angles of the three guide lights;

[0022] 3) The pose calculation network calculates the coordinate values; the trained AUV 3D pose calculation network is used to calculate the AUV's pose; the input to the AUV 3D pose calculation network is... The output is the coordinate position corresponding to that angle. .

[0023] Beneficial effects

[0024] This invention proposes a highly reliable intelligent optical positioning method for underwater unmanned platforms. By establishing a three-dimensional pose angle-coordinate dataset and a three-dimensional pose calculation network model for AUVs, and designing a geometric loss function for AUV three-dimensional pose calculation, fast and accurate AUV three-dimensional pose calculation is achieved. The angle error of the model pose calculation is within... This method meets the positioning accuracy requirements for underwater unmanned platforms during docking. The hardware relied upon is a multi-quadrant sensor, which has advantages such as small size, small data volume, and simple installation. During deployment, the method also boasts advantages such as low hardware computation and rapid pose calculation, with a computational load of only 894.0 Mac and a parameter count of only 831. Attached Figure Description

[0025] Figure 1 It is a coordinate diagram of the beacon system.

[0026] Figure 2 This is a diagram showing the three-dimensional pose angle relationship of an AUV.

[0027] Figure 3 This is a flowchart of the process for establishing an AUV 3D pose angle-coordinate dataset according to the present invention.

[0028] Figure 4 This is the training flowchart of the present invention.

[0029] Figure 5 This is a flowchart of the testing process of the present invention. Detailed Implementation

[0030] The present invention will be further described first with reference to the embodiments and accompanying drawings:

[0031] The high-reliability intelligent optical positioning method for underwater unmanned platforms in this embodiment includes the following steps:

[0032] First: Establish an AUV 3D pose angle-coordinate dataset, see [reference]. Figure 3 This includes the following sub-steps

[0033] Step 1: Establish a three-dimensional coordinate system between the AUV and the recovery dock. For example... Figure 1 As shown.

[0034] Step 2: Randomly and uniformly sample to obtain coordinate data. Define the AUV coordinates. The coordinate range, denoted as For the farthest distance of guidance, then , , Within the coordinate range of the AUV, a coordinate value is obtained by random and uniform sampling. .

[0035] Step 3: Calculate the angle value corresponding to the coordinate value. , , Solving for , , At this point, a set of three-dimensional pose angle-coordinate data of the AUV is obtained. See Figure 2 .

[0036] Secondly: Train the AUV 3D pose calculation network model, such as Figure 4 As shown, it includes the following sub-steps:

[0037] Step 4: Create an AUV 3D pose angle-coordinate dataset. Repeat steps 2 and 3 a total of 10 million times to generate 10 million sets of angle-coordinate data.

[0038] Step 5: Establish the AUV 3D pose calculation network model. The AUV 3D pose calculation network model is a seven-layer perceptron network, with the number of neurons in each layer being as follows: Except for the last layer This applies to all layers except the activation function. Activation function.

[0039] Step 6: Train the AUV 3D pose calculation network model. (Using...) Optimize the algorithm, set the learning rate to The input to the AUV 3D pose calculation network is the angle data from the set of angle-coordinate data in step 4. The output is the calculated coordinate position. .

[0040] Step 7: Define the geometric loss function for AUV 3D pose resolution. The geometric loss function for AUV 3D pose resolution is the calculated coordinate position... With coordinate position The mean squared error of the geometric relationship. Wherein, The corresponding angular coordinates are calculated from step 3. , This represents the distance between the AUV coordinates calculated by the network and the origin of the beacon system. Represents the coordinate position in the dataset The distance from the origin of the beacon system. The geometric loss function for AUV 3D pose calculation is specifically defined as:

[0041]

[0042] in, , , , , , .

[0043] Step 7: Optimize the AUV 3D pose calculation network model parameters. For each optimization, 64 data points are randomly selected from the data in Step 3. Following the optimization method defined in Step 5 and the loss function defined in Step 6, the AUV 3D pose calculation network model parameters from Step 5 are optimized 10 million times using backpropagation. The optimal AUV 3D pose calculation network model is obtained.

[0044] Finally, the hardware is deployed to calculate the AUV's 3D pose, such as... Figure 5 As shown, it includes the following sub-steps:

[0045] Step 8: Calculate the deflection angle of the guide lights in the beacon system using a multi-quadrant sensor. The multi-quadrant sensor detects each guide light and obtains a set of voltage values. The horizontal deflection angle of the guide light is determined by the following formula. with vertical deflection angle Angle:

[0046] ,

[0047] The horizontal and vertical deflection angles of the three guide lights are denoted as follows: , , , , , .

[0048] Step 9: Calculate the angles between any two guide lights and the center of the multi-quadrant detector. The formula is:

[0049] ,

[0050] The angles obtained are as follows: .

[0051] Step 10: The AUV 3D pose calculation network model calculates the AUV coordinates. The included angle obtained in Step 9 is used as the reference. As input to the AUV 3D pose calculation network model obtained in step 7, the output is the AUV coordinates. .

Claims

1. A highly reliable intelligent positioning method for underwater unmanned platforms, characterized in that, The method is based on the following hardware setup: a beacon system consisting of three light sources is arranged in an equilateral triangle at the docking station's return port; a large-view area array sensor is positioned at the head of the AUV to receive the light intensity signal from the docking station's beacon system; the method includes the following processes: First: Establish an AUV 3D pose angle-coordinate dataset, including the following sub-steps: 1) Establish the beacon system coordinate system; the beacon system is located on the recovery dock, and the three guiding light sources are arranged in an equilateral triangle with a side length of [missing information]. The divergence angle of the guiding light source is A rectangular coordinate system is established using the incenter of an equilateral triangle as the origin of the docking station coordinate system; the coordinates of the three guiding light sources are respectively... , , ; 2) Calculate AUV angle data based on AUV coordinate data: Assume the AUV is located at coordinate... At this point, the vector , , Solving for , , ; 3) Establish an AUV 3D pose angle-coordinate dataset: constrain AUV coordinates The coordinate range, denoted as For the farthest distance of guidance, then , , Within the coordinate range of the AUV, uniform sampling is used to obtain the value of a coordinate. The corresponding angle data is calculated from step 2). At this point, a set of three-dimensional pose angle-coordinate data of the AUV is obtained. ; Then: Train the AUV 3D pose calculation network model, including the following sub-steps: 1) Establish an AUV 3D pose calculation network model; the AUV 3D pose calculation network model is a multilayer perceptron network or a Transformer regression network model; 2) Train the AUV 3D pose solving network. The input of the AUV 3D pose solving network is the angle information in the AUV 3D pose angle-coordinate dataset, and the output is the solved coordinate position. The loss function is the mean square error of the geometric relationship between the solved coordinate position and the actual coordinate position. Use deep learning algorithms to optimize the network parameters and save the trained model parameters. Finally: Hardware deployment for calculating the AUV's 3D pose includes the following sub-steps: 1) The multi-quadrant detector calculates the guide light angle of the beacon system; the voltage generated by the guide light source in the beacon system in the multi-quadrant detector of the AUV head is... The horizontal deflection angle is determined by the following formula. with vertical deflection angle Angle: ； in, , , , For multi-quadrant sensors, the calibration value is a constant; 2) Calculate the angle between any two guide lights and the center of the multi-quadrant detector; the angle is calculated using the following formula: , , , , , The horizontal and vertical deflection angles of the three guide lights; 3) The pose calculation network calculates the coordinate values; the trained AUV 3D pose calculation network is used to calculate the AUV's pose; the input to the AUV 3D pose calculation network is... The output is the coordinate position corresponding to that angle. .

Images