Fast crawling control method for biomimetic adsorbing underwater robot

Abstract

The application relates to the technical field of three-dimensional position control, and provides a quick crawling control method for a bionic adsorption underwater robot, which comprises the following steps: collecting posture information and multi-dimensional environmental data of the underwater robot, and obtaining a three-dimensional grid map of an underwater surface; obtaining a planned path of the underwater robot; dividing the planned path into a plurality of small path segments; obtaining a flow prediction time length at a current time, and obtaining a flow prediction vector at each time within the flow prediction time length; obtaining a speed space of the underwater robot at the current time; constructing a robot crawling prediction model; obtaining a maximum reachable speed at each time within the flow prediction time length; screening a plurality of initial paths at the current time; obtaining a best path at the current time; and controlling the bionic adsorption underwater robot to quickly crawl. The dynamic interference of the environment is corrected, the speed space of a traditional algorithm is dynamically adjusted, and the quick crawling demand in a complex underwater environment is met.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional position control technology, and more specifically to a rapid crawling control method for a biomimetic adsorption underwater robot. Background Technology

[0002] Underwater robots, as important equipment in modern underwater operations, are widely used in fields such as marine resource development, environmental monitoring, facility maintenance, and underwater search and rescue. Among them, biomimetic adsorption underwater robots stand out in complex underwater environments due to their flexible movement capabilities and stable adsorption performance. Their tracked or multi-legged biomimetic design enables them to perform high-precision inspection, repair, and cleaning tasks on irregular surfaces such as ship hulls, pipelines, and bridge piers.

[0003] However, in practical applications, the complexity of the underwater environment poses a severe challenge to robots. Dynamic factors such as water flow and turbulence, combined with static characteristics such as surface roughness, pressure depth, and temperature changes, make the motion path planning and control of underwater robots more difficult. This complexity requires path planning algorithms to consider the robot's own performance (such as motor speed and suction force range) and obstacle distribution, while also being able to respond quickly to dynamic changes in the environment to ensure that the robot can achieve accurate and rapid crawling.

[0004] Existing local path planning algorithms, such as the Dynamic Window (DWA) algorithm, are widely used in underwater robot motion planning due to their simplified model assumptions and efficient computation. However, traditional DWA algorithms mainly rely on the robot's own motion constraints and obstacle information in the static environment for path planning, which is insufficient for adapting to complex dynamic underwater environments. Especially when there is significant interference from water flow turbulence, large changes in surface friction coefficient, or pressure differences affecting motion parameters, the motion often deviates from the expected path, reducing the robot's crawling speed and stability. Summary of the Invention

[0005] This invention provides a rapid crawling control method for a biomimetic adsorption underwater robot, to solve the problem of existing methods interfering with the rapid crawling speed control of robots in complex underwater environments. The specific technical solution adopted is as follows:

[0006] This invention proposes a rapid crawling control method for a biomimetic adsorption underwater robot, which includes the following steps:

[0007] Collect attitude information and multi-dimensional environmental data of the underwater robot, and obtain a three-dimensional grid map of the underwater surface;

[0008] Obtain the planned path of the underwater robot; based on the multi-dimensional environmental data of each grid in the planned path, divide the planned path into several small path segments; based on the differences between the multi-dimensional environmental data at different times in the same small path segment, obtain the historical water flow stability at each time in each small path segment, and then obtain the water flow prediction duration at the current time, and obtain the water flow prediction vector at each time within the water flow prediction duration.

[0009] Obtain the velocity space of the underwater robot at the current moment; construct a robot crawling prediction model based on the robot's attitude information at each moment and multi-dimensional environmental data; obtain the maximum achievable velocity at each moment within the water flow prediction time through the robot crawling prediction model, based on the water flow prediction vector, the planned path, and multi-dimensional environmental data at each position in the planned path; and select several initial paths at the current moment by combining the velocity space.

[0010] The optimal path at the current moment is obtained by using the evaluation function of the DWA algorithm for several initial paths, and the biomimetic adsorption underwater robot is controlled to crawl quickly.

[0011] Optionally, the planned path of the underwater robot is obtained by the following method:

[0012] Based on a 3D grid map of the underwater surface, the planned path of the underwater robot is obtained using the A-star algorithm.

[0013] Optionally, the specific method for dividing the planned path into several smaller path segments based on the multi-dimensional environmental data of each grid in the planned path includes:

[0014] Based on the multi-dimensional environmental data of each grid in the planning path, the segmentation probability of each grid in the planning path is obtained; a coordinate system is constructed with the grid order value as the x-axis and the segmentation probability as the y-axis to obtain the segmentation probability change curve of the planning path.

[0015] Obtain several maxima in the segmented probability change curve, take the grid corresponding to each maxima as the segment point of the planned path, take the first grid in the planned path as a segment point, and divide the planned path into several small segments through the segment points.

[0016] Optionally, the segmentation probability of each grid cell in the planned path is obtained using the following method:

[0017] Each grid cell in the 3D raster map corresponds to a local normal vector, tilt angle, curvature, and underwater depth, with tilt angle, curvature, and underwater depth serving as the data information dimensions for each grid cell; in the planned path, the first... The segmentation probability of each grid cell The calculation method is as follows:

[0018]

[0019] in, Indicates the number of dimensions of data information; Indicates the preset reference range; This indicates that in the planned path, starting from the first... The grid to the first The first grid cell The average environmental data across each data information dimension; This indicates that in the planned path, starting from the first... The grid to the first The first grid cell The average environmental data across each data information dimension; Indicates the first path in the planned path The first grid cell Environmental data in multiple dimensions; This represents the function for calculating the standard deviation. This represents the absolute value function.

[0020] Optionally, the historical water flow stability at each moment in each short path segment is obtained using the following method:

[0021] Based on the differences in multi-dimensional environmental data between historical and current locations within the same short path segment, the underwater surface similarity between each historical and current location is obtained.

[0022] For the current segment of the planned path, based on the water flow velocity and direction at the corresponding position of the underwater robot within that segment at the current time and at each historical time, the water flow vectors for the current time and each historical time are constructed; the historical water flow stability at the current time... The calculation method is as follows:

[0023]

[0024] in, This indicates the current time within this short path segment. The number of previous historical moments; Representing historical moments With the current moment underwater surface similarity; Representing historical moments The water flow vector, Indicates the current time The water flow vector; This represents the cosine similarity function for vectors.

[0025] Optionally, the underwater surface similarity between each historical moment and the current moment is obtained using the following method:

[0026] Obtain multi-dimensional environmental data at the current location and construct the current time. underwater surface data vector ,in Indicates the current time The local normal vector at the corresponding position, Indicates the current time The tilt angle at the corresponding position Indicates the current time Curvature at the corresponding position Indicates the current time The underwater depth at the corresponding location;

[0027] Get historical moments under this short path segment underwater surface data vector ,in Representing historical moments The local normal vector at the corresponding position, Representing historical moments The tilt angle at the corresponding position Representing historical moments Curvature at the corresponding position Representing historical moments The underwater depth at the corresponding location;

[0028] Historical moment With the current moment underwater surface similarity The calculation method is as follows:

[0029]

[0030] in, Representing historical moments underwater surface data vectors, Indicates the current time underwater surface data vector; This indicates the calculation of the Euclidean norm of a vector; To avoid hyperparameters with a denominator of 0; This represents the normalization function.

[0031] Optionally, the specific method for obtaining the water flow prediction duration at the current moment includes:

[0032]

[0033] in, Indicates the current time Water flow prediction duration Indicates the current time Historical water flow stability and These represent the minimum and maximum prediction durations in the DWA algorithm, respectively.

[0034] Optionally, the specific method for constructing the robot crawling prediction model includes:

[0035] The input layer is constructed using all moments of the underwater robot up to the current moment, including the water flow vector at each moment, the underwater depth of the robot at the corresponding position at each moment, the robot's crawling direction and the attitude vector composed of the robot's attitude information, as well as several motor powers.

[0036] The hidden layers of a multilayer perceptron are constructed and activation functions are set; the output layer consists of two neurons, corresponding to linear velocity and angular velocity respectively; through construction and training, the trained multilayer perceptron is finally obtained and used as a robot crawling prediction model.

[0037] Optionally, based on the water flow prediction vector, the planned path, and multi-dimensional environmental data at each location along the planned path, a robot crawling prediction model is used to obtain the maximum achievable speed at each moment within the water flow prediction time; combined with the speed space, several initial paths are selected for the current moment. The specific methods include:

[0038] Linear velocity is sampled at a certain sampling rate in the velocity space. With angular velocity Data is collected separately to obtain several linear velocities and several angular velocities. Any linear velocity and any angular velocity are combined to obtain several sets of velocities, and each set of velocities corresponds to a segment of the predicted path.

[0039] For any given moment within the current flow prediction time, based on the flow prediction vector at that moment, for any segment of the prediction path, there is a corresponding set of linear and angular velocities. The position at that moment within the predicted path segment is obtained, along with the underwater depth and the robot's crawling direction at that position. Simultaneously, based on the local normal vector of the grid at that position, the robot's attitude vector is obtained and input into the robot crawling prediction model. The motor power is set to its maximum value, and the maximum achievable linear velocity and maximum achievable angular velocity at that moment within the predicted path segment are output. The maximum achievable linear velocity and maximum achievable angular velocity at each moment within the flow prediction time within the predicted path segment are also obtained.

[0040] Under this predicted path segment, if, within a certain time period of the predicted water flow duration, the maximum achievable linear velocity is less than the linear velocity corresponding to this predicted path segment, or the maximum achievable angular velocity is less than the angular velocity corresponding to this predicted path segment, then this predicted path segment is deleted. The above judgment is performed on the predicted paths corresponding to each group of velocities in the velocity space, and the remaining predicted paths are used as several initial paths.

[0041] Optionally, the specific method for obtaining the optimal path at the current moment from several initial paths using the evaluation function of the DWA algorithm includes:

[0042] By using distance evaluation coefficient, azimuth evaluation coefficient, and velocity evaluation coefficient, and based on each initial path and a set of linear and angular velocities in the corresponding velocity space, a score is obtained for each initial path, and the initial path with the highest score is taken as the optimal path.

[0043] The beneficial effects of this invention are as follows: This invention collects the attitude information and multi-dimensional environmental data of an underwater robot, and obtains a three-dimensional grid map of the underwater surface based on an underwater three-dimensional model; it obtains a planned path based on the three-dimensional grid map of the underwater surface, and segments the path according to changes in the underwater surface environment. By analyzing the similarity of the underwater surface environment and water flow changes within the current short path segment, it comprehensively quantifies the historical water flow stability under the current short path segment, and determines the water flow prediction duration accordingly, thereby obtaining the water flow prediction vector at several moments within the water flow prediction duration; based on the traditional DWA algorithm for constructing a velocity space... By using multi-dimensional environmental data corresponding to the crawled locations and the robot's posture information at the corresponding moments, a robot crawling prediction model is constructed. Based on the speed performance of the predicted paths under the robot crawling prediction model for several groups of speeds in the velocity space, an initial path is selected to ensure that the corresponding speed can resist the influence of the environment, thereby completing the planned path crawling. By quantifying the complex environmental changes during the underwater robot's crawling process, the influence of multi-dimensional environmental data on the robot's crawling speed and direction is analyzed, thereby optimizing the traditional path planning algorithm and improving the anti-interference ability of the underwater robot during the crawling process. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1This is a schematic flowchart of a rapid crawling control method for a biomimetic adsorption underwater robot provided in one embodiment of the present invention;

[0046] Figure 2 A schematic diagram of the crawling control architecture for an underwater robot;

[0047] Figure 3 This is a schematic diagram of an underwater 3D model. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Please see Figure 1 The diagram illustrates a flowchart of a rapid crawling control method for a biomimetic adsorption underwater robot according to an embodiment of the present invention. The method includes the following steps:

[0050] Step S001: Collect the attitude information and multi-dimensional environmental data of the underwater robot, and obtain a three-dimensional grid map of the underwater surface.

[0051] The purpose of this embodiment is to comprehensively control the rapid crawling of the biomimetic adsorption underwater robot by using the underwater robot's posture information and the complex environmental data collected by sensors. This requires collecting the underwater robot's posture information and multi-dimensional environmental data, and simultaneously constructing a three-dimensional model of the underwater environment.

[0052] Specifically, the underwater robot's attitude information, including pitch, roll, and yaw angles, is acquired through an inertial measurement unit (IMU); the IMU model used is XSENS MTi-30 / MTi-300; the acceleration range in the inertial measurement unit parameter configuration is as follows: arrive Angular velocity range is arrive The update rate is 100Hz to 1kHz, and the temperature compensation is -40℃ to 85℃. The inertial measurement unit is installed near the robot's center of gravity to reduce motion coupling errors.

[0053] Furthermore, the underwater robot's depth during its movement is collected using a water pressure sensor. The water pressure sensor model is Keller Series 33X / 35X (high precision); its parameters specify an operating range of 0 to 10 bar (suitable for shallow water) and an accuracy of [missing information]. (Full Scale) The output signal is an analog signal or a digital signal; the water pressure sensor is installed on the bottom of the robot, in contact with the water environment, and the underwater depth is estimated by measuring the water pressure.

[0054] Furthermore, point cloud data of the underwater environment is collected using sonar for navigation, obstacle avoidance, and target identification. The sonar model used is the Imagenex 881A, a multi-beam sonar. The sonar's parameters include a frequency range of 310kHz to 1MHz, a maximum ranging range of 50m to 200m, and a fan-shaped scanning range of [missing information]. arrive It is mounted on the front of the robot for navigation, obstacle detection, and obstacle avoidance.

[0055] Furthermore, water flow data, including velocity and direction, is collected using a flow meter; the flow meter model used is the SonTek Argonaut ADV, an acoustic Doppler flow meter; the flow velocity range is configured in the flow meter parameters. arrive Accuracy is The operating frequency is 1MHz to 3MHz; it is installed on the bottom of the robot to avoid interference from the robot's movement and to ensure that the flow field at the sensor measurement position is uniform.

[0056] Furthermore, after sonar acquires point cloud data of the underwater environment, such as Figure 3 The diagram shows a schematic of an underwater 3D model. By converting the point cloud data into a 3D raster map through meshing, a 3D raster map of the underwater surface is obtained. Each grid cell in the 3D raster map corresponds to a local normal vector, tilt angle, curvature, and underwater depth.

[0057] Step S002: Obtain the planned path of the underwater robot; based on the multi-dimensional environmental data of each grid in the planned path, divide the planned path into several small path segments; based on the differences between the multi-dimensional environmental data at different times in the same small path segment, obtain the historical water flow stability at each time in each small path segment, and then obtain the water flow prediction duration at the current time, and obtain the water flow prediction vector at each time within the water flow prediction duration.

[0058] Preferably, in one embodiment of the present invention, the specific method for obtaining the planned path of the underwater robot includes:

[0059] After obtaining the three-dimensional grid map of the underwater surface, the planned path of the underwater robot is obtained through the traditional A-star algorithm. The planned path includes the path that the underwater robot has already traveled and the subsequent predicted path. The A-star algorithm is an existing method for shortest path search, and will not be described in detail in this embodiment.

[0060] It should be noted that the requirements for tilt angle, curvature, and adhesion force vary in different regions, necessitating segmentation of the planned path. Each segment employs a dynamic adjustment strategy, optimizing crawling speed and local paths to adapt to different surface characteristics. The DWA algorithm excels at real-time optimization of local paths, grouping regions with similar environmental characteristics into the same segment. This allows the robot to adopt consistent speed adjustment and attitude control strategies, resulting in minimal changes to the motion model and control parameters, improved operational efficiency, and more coherent path planning and execution.

[0061] Preferably, in one embodiment of the present invention, the planned path is divided into several smaller path segments based on the multi-dimensional environmental data of each grid in the planned path, including the following specific method:

[0062] In the 3D grid map of the underwater surface, each grid cell collects multi-dimensional environmental data, including the corresponding local normal vector, tilt angle, curvature, and underwater depth. The tilt angle, curvature, and underwater depth are used as the data information dimensions for each grid cell. Then, in the planned path, the... The segmentation probability of each grid cell The calculation method is as follows:

[0063]

[0064] in, This represents the number of dimensions of the data information, in this embodiment. ; This indicates a preset reference range, which is adopted in this embodiment. To narrate; This indicates that in the planned path, starting from the first... The grid to the first The first grid cell The average environmental data across each data information dimension; This indicates that in the planned path, starting from the first... The grid to the first The first grid cell The average environmental data across each data information dimension; Indicates the first path in the planned path The first grid cell Environmental data in multiple dimensions; This represents the function for calculating the standard deviation. This represents an absolute value function. It should be noted that if the number of grids before or after any grid is less than the preset reference range, the reference range on the corresponding side is constructed with the actual existing grids, and the same number of grids are also used to construct the reference range on the other side.

[0065] It should be noted that in the planning path, the greater the change in environmental data dimensions before and after the grid, that is, the greater the change in tilt angle, curvature and underwater depth, the greater the difference in the environment of the path before and after the grid, the greater the possibility of segmentation, and the greater the probability of grid segmentation.

[0066] Furthermore, following the method described above, the segmentation probability of each grid cell in the planned path is obtained. A coordinate system is constructed with the grid order value as the abscissa and the segmentation probability as the ordinate to obtain the segmentation probability variation curve of the planned path. Several maxima in the segmentation probability variation curve are obtained, and the grid cells corresponding to each maxima are taken as segmentation points of the planned path. The first grid cell in the planned path is also taken as a segmentation point. The planned path is divided into several small path segments through the segmentation points, where each segmentation point is the first grid cell in a small path segment.

[0067] Preferably, in one embodiment of the present invention, based on the differences between multi-dimensional environmental data at different times and locations within the same short path segment, the historical water flow stability at each time point within each short path segment is obtained, thereby obtaining the water flow prediction duration at the current time, and obtaining the water flow prediction vector at each time point within the water flow prediction duration. The specific method includes:

[0068] It should be noted that after the planned path is divided into small segments, the local path needs to be planned in real time using the DWA algorithm within each small segment. When the underwater robot moves within a small segment, the direction of the underwater current at different times will affect the robot's crawling speed. Therefore, it is necessary to predict the subsequent water flow speed based on the changes in the water flow at historical times.

[0069] It should be further noted that in order to predict and analyze the changes in water flow velocity and direction, it is necessary to consider whether the water flow changes are stable over a period of time in history. However, the robot's position is not constant at any given time. If the robot has crawled a long distance over a period of time, the underwater structure may have changed significantly. In this case, the accuracy of predictions for future times based on the water flow velocity at a given time in history will be reduced. Therefore, it is necessary to analyze the stability of the water flow at a given time in history and consider the similarity between the underwater path surface features at the robot's position at a given time in history and those at the current time.

[0070] Specifically, for the short path segment in the planned path at the current moment, obtain multi-dimensional environmental data at the corresponding location at the current moment to construct the current moment's path. underwater surface data vector ,in Indicates the current time The local normal vector at the corresponding position, Indicates the current time The tilt angle at the corresponding position Indicates the current time Curvature at the corresponding position Indicates the current time The underwater depth at the corresponding location; similarly, the historical moments for this short path segment can be obtained. underwater surface data vector ,in Representing historical moments The local normal vector at the corresponding position, Representing historical moments The tilt angle at the corresponding position Representing historical moments Curvature at the corresponding position Representing historical moments The corresponding underwater depth; it should be noted that in the underwater surface data vectors of historical and current times, the local normal vector, tilt angle, and curvature are all obtained based on a grid, while the underwater depth is estimated by a water pressure sensor deployed on the bottom of the robot; therefore, the historical time... With the current moment underwater surface similarity The calculation method is as follows:

[0071]

[0072] in, Representing historical moments underwater surface data vectors, Indicates the current time underwater surface data vector; This indicates the calculation of the Euclidean norm of a vector; To avoid hyperparameters with a denominator of 0, this embodiment adopts... To narrate; This represents a linear normalization function, which normalizes the relationship between each historical moment and the current moment within this short path segment. .

[0073] It should be noted that the similarity of underwater surfaces is quantified by the difference between the underwater surface data vectors at historical and current times. The smaller the difference between the underwater surface data vectors, the more similar the underwater surface environments are, and the greater the underwater surface similarity is.

[0074] Furthermore, since the current meter can measure the velocity and direction of underwater water flow, the underwater robot can obtain the water flow velocity and direction at the corresponding location within this short path segment at both the current moment and each historical moment through the current meter, thus constructing the water flow vectors at the current moment and each historical moment; therefore, the historical water flow stability at the current moment can be determined. The calculation method is as follows:

[0075]

[0076] in, This indicates the current time within this short path segment. The number of previous historical moments; Representing historical moments With the current moment underwater surface similarity; Representing historical moments The water flow vector, Indicates the current time The water flow vector; This represents the cosine similarity function for vectors.

[0077] It should be noted that the similarity of the current flow vectors with historical times is quantified. The more similar the flow vectors are, the more stable the flow behavior within a short path segment. At the same time, the similarity of the underwater surface is used as a weight to analyze the similarity of the flow at different underwater locations, and then the overall flow stability during the crawling of this short path segment is comprehensively quantified, that is, the historical flow stability at the current time. After obtaining the historical flow stability, the flow prediction duration at the current time needs to be quantified based on the time period predicted backward by the traditional DWA algorithm. Based on the maximum and minimum values ​​of the traditional predicted time period length, the more stable the flow within the short path segment, that is, the greater the historical flow stability, the more the maximum value of the predicted duration is referred, and vice versa.

[0078] Furthermore, at the current moment Water flow prediction duration The calculation method is as follows:

[0079]

[0080] in, Indicates the current time Historical water flow stability and These represent the minimum and maximum prediction durations in the DWA algorithm, respectively. This embodiment uses... and To narrate.

[0081] Furthermore, in this embodiment, the update time interval of the DWA algorithm is set to 20ms. Then, within the current water flow prediction time, several water flow prediction vectors are obtained based on the water flow vector using the traditional exponential moving weighted average algorithm. These are the water flow prediction vectors at several moments within the water flow prediction time. The exponential moving weighted average algorithm is an existing algorithm, and in this embodiment, the exponent is described as 0.9.

[0082] Thus, the planned path is obtained based on the three-dimensional grid map of the underwater surface, and segmented according to the changes in the underwater surface environment. By analyzing the similarity of the underwater surface environment and water flow changes within the current small segment of the path, the historical water flow stability under the current small segment of the path is comprehensively quantified, and the water flow prediction duration is determined accordingly. In this way, the water flow prediction vector at several moments within the water flow prediction duration is obtained.

[0083] Step S003: Obtain the velocity space of the underwater robot at the current moment; construct a robot crawling prediction model based on the posture information of the underwater robot at each moment and multi-dimensional environmental data; obtain the maximum achievable velocity at each moment within the water flow prediction time through the robot crawling prediction model based on the water flow prediction vector, the planned path and multi-dimensional environmental data at each position in the planned path; and select several initial paths at the current moment by combining the velocity space.

[0084] Preferably, in one embodiment of the present invention, the specific method for obtaining the velocity space of the underwater robot at the current moment includes:

[0085] It should be noted that within a single short path segment, the traditional DWA algorithm is needed to adjust the speed. However, the traditional speed space only considers the robot's motor performance parameters and obstacle conditions, without taking into account the complex underwater current environment and the differences in the robot's crawling speed under different postures. This leads to a deviation between the predicted path and the actual crawling path. For example, when the robot's direction of travel is opposite to the direction of the current, the robot's maximum achievable speed will be significantly reduced. When the planned speed value cannot be actually reached, the speed space will fail, and the robot will not be able to reach the target position through the expected trajectory.

[0086] Specifically, using the traditional DWA algorithm, based on velocity boundary constraints and the influence of motor performance and obstacles on the underwater robot, the velocity space of the underwater robot at the current moment is obtained. Specifically:

[0087]

[0088] in, Linear velocity, Angular velocity, and These represent the lower and upper limits of the linear velocity, respectively. and These represent the lower and upper limits of angular velocity, respectively.

[0089] It should be noted that the traditional DWA algorithm for obtaining the velocity space is an existing technology and will not be described in detail in this embodiment; while the velocity boundary limit, the influence of the underwater robot on the motor performance and the influence of obstacles are all parameters that are directly obtained, and the specific methods for obtaining them will not be described in this embodiment.

[0090] Preferably, in one embodiment of the present invention, a robot crawling prediction model is constructed based on the underwater robot's attitude information at various times and multi-dimensional environmental data, including the following specific methods:

[0091] It should be noted that for underwater robots, different water flow vectors and crawling directions will have different effects on the robot's crawling speed. Therefore, it is necessary to obtain the influence relationship of multiple factors on the robot's linear velocity and angular velocity. By constructing and training a multilayer perceptron, a robot crawling prediction model can be obtained.

[0092] Specifically, the input layer is constructed using all moments of the underwater robot before the current moment (including moments within other short path segments before), including the water flow vector at each moment, the underwater depth of the robot at the corresponding position at each moment, the robot's crawling direction and the attitude vector composed of the robot's attitude information, as well as several motor power values.

[0093] For the hidden layers of the multilayer perceptron (MLP), this embodiment is designed with 3 layers, each with 100 neurons, using ReLU as the activation function; the output layer has two neurons, corresponding to linear velocity and angular velocity respectively; through construction and training, the trained multilayer perceptron is finally obtained and used as a robot crawling prediction model.

[0094] Preferably, in one embodiment of the present invention, based on the water flow prediction vector, the planned path, and multi-dimensional environmental data at each location along the planned path, the maximum achievable speed at each moment within the water flow prediction time is obtained through a robot crawling prediction model; combined with the speed space, several initial paths are selected for the current moment, including the following specific method:

[0095] Get the current time After determining the velocity space, the linear velocity is sampled at a certain sampling rate within that velocity space. With angular velocity Data was collected separately, with the linear velocity being... angular velocity Data is collected to obtain several linear velocities and several angular velocities. Any linear velocity and any angular velocity are combined to obtain several sets of velocities. In the DWA algorithm, each set of velocities corresponds to a segment of the predicted path.

[0096] It should be noted that in velocity space, there exists a set of velocities at which the predicted path remains a straight line. When the linear velocity and angular velocity change, the predicted path will deviate.

[0097] Furthermore, for any moment within the current water flow prediction time, based on the water flow prediction vector at that moment, for any segment of the prediction path, there is a corresponding set of linear velocity and angular velocity. By obtaining the position corresponding to that moment within that segment of the prediction path, the underwater depth at that moment and the robot's crawling direction at that position are obtained. Simultaneously, based on the local normal vector of the grid at that position, the robot's attitude vector at that position is obtained and input into the robot crawling prediction model. With the motor power set to the maximum achievable value, the maximum achievable linear velocity and maximum achievable angular velocity at that moment within that segment of the prediction path can be output. The maximum achievable linear velocity and maximum achievable angular velocity at each moment within the water flow prediction time within that segment of the path are also obtained.

[0098] Furthermore, under this predicted path segment, if within a certain time period of the predicted water flow duration, there exists a moment where the maximum achievable linear velocity is less than the linear velocity corresponding to this predicted path segment, or the maximum achievable angular velocity is less than the angular velocity corresponding to this predicted path segment, then this predicted path segment is deleted; the above judgment is performed on the predicted paths corresponding to each group of velocities in the velocity space, and the remaining predicted paths segment is used as several initial paths.

[0099] It should be noted that if the maximum reachable linear velocity or maximum reachable angular velocity is less than the velocity of the corresponding group under the corresponding predicted path, the crawling cannot be completed normally according to the predicted path. That is, the planning fails due to environmental influences. In this case, such predicted paths need to be removed to avoid environmental interference, and predicted paths with strong resistance to environmental interference are retained as the initial paths.

[0100] Thus, based on the velocity space constructed by the traditional DWA algorithm, a robot crawling prediction model is constructed by using multi-dimensional environmental data corresponding to the crawled locations and the robot's posture information at the corresponding time. Based on the velocity performance of the paths predicted by several groups of velocities in the velocity space under the robot crawling prediction model, the initial path is selected to ensure that the corresponding group of velocities can resist the influence of the environment, thereby completing the planned path crawling.

[0101] Step S004: Obtain the optimal path at the current moment from several initial paths using the evaluation function of the DWA algorithm, and control the biomimetic adsorption underwater robot to crawl quickly.

[0102] It should be noted that after the initial path is obtained, the initial path can meet the linear velocity and angular velocity required by the corresponding planned path. In the process of planning the path, the DWA algorithm uses the evaluation function to comprehensively quantify the selectivity of the initial path through distance evaluation, azimuth evaluation and velocity evaluation, so as to determine the optimal path.

[0103] Specifically, each initial path corresponds to a set of linear and angular velocities in the underwater robot's velocity space at the current moment. The evaluation function of the DWA algorithm obtains the evaluation score of each initial path through distance evaluation coefficient, azimuth evaluation coefficient, and velocity evaluation coefficient. In this embodiment, the distance evaluation coefficient is described as 3, the azimuth evaluation coefficient as 2, and the velocity evaluation coefficient as 1.5. Based on each initial path and its corresponding set of linear and angular velocities in the velocity space, the score of each initial path is obtained, and the initial path with the highest score is taken as the optimal path.

[0104] Furthermore, after obtaining the optimal path at the current moment, the underwater robot acquires the subsequent route and speed parameters, namely a set of linear and angular velocities corresponding to the optimal path. The optimal path includes the optimal linear and angular velocity commands for the corresponding time period. Then, using a traditional PID control algorithm, the robot's optimal linear and angular velocities are adjusted by calculating the error between the underwater robot's current speed state and the target state (optimal linear and angular velocities), thereby controlling the underwater robot's subsequent rapid crawling. The time period corresponding to the optimal path is the water flow prediction duration at the current moment. After the crawling ends within the water flow prediction duration, the optimal path is re-obtained using the above method to achieve rapid crawling control of the biomimetic adsorption underwater robot.

[0105] Thus, by quantifying the complex environmental changes during the underwater robot's crawling process and analyzing the impact of multi-dimensional environmental data on the robot's crawling speed and direction, the traditional path planning algorithm is optimized, thereby improving the underwater robot's anti-interference ability during the crawling process.

[0106] Please see Figure 2 The diagram illustrates the architecture of an underwater robot crawling control system, used to achieve rapid crawling control for a biomimetic adsorption underwater robot. After the sensors collect data, the main control chip processes the aforementioned rapid crawling control method for the biomimetic adsorption underwater robot and then outputs control commands. The PID control algorithm outputs commands to adjust the adsorption control module and the motion control module, respectively controlling the robot's adsorption and crawling. The main control chip uses an STM32, which is the most common general-purpose chip, suitable for robots with moderate task complexity, and is used for low power consumption, real-time control, and sensor data processing.

[0107] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A rapid crawling control method for a biomimetic adsorption underwater robot, characterized in that, The method includes the following steps: Collect attitude information and multi-dimensional environmental data of the underwater robot, and obtain a three-dimensional grid map of the underwater surface; Obtain the planned path of the underwater robot; based on the multi-dimensional environmental data of each grid in the planned path, divide the planned path into several small path segments; based on the differences between the multi-dimensional environmental data at different times in the same small path segment, obtain the historical water flow stability at each time in each small path segment, and then obtain the water flow prediction duration at the current time, and obtain the water flow prediction vector at each time within the water flow prediction duration. Obtain the velocity space of the underwater robot at the current moment; construct a robot crawling prediction model based on the robot's attitude information at each moment and multi-dimensional environmental data; obtain the maximum achievable velocity at each moment within the water flow prediction time through the robot crawling prediction model, based on the water flow prediction vector, the planned path, and multi-dimensional environmental data at each position in the planned path; and select several initial paths at the current moment by combining the velocity space. The optimal path at the current moment is obtained by using the evaluation function of the DWA algorithm for several initial paths, and the biomimetic adsorption underwater robot is controlled to crawl quickly. The specific method for obtaining the historical water flow stability at each moment in each small path segment is as follows: Based on the differences in multi-dimensional environmental data between historical and current locations within the same short path segment, the underwater surface similarity between each historical and current location is obtained. For the current segment of the planned path, based on the water flow velocity and direction at the corresponding position of the underwater robot within that segment at the current time and at each historical time, the water flow vectors for the current time and each historical time are constructed; the historical water flow stability at the current time... The calculation method is as follows: in, This indicates the current time within this short path segment. The number of previous historical moments; Representing historical moments With the current moment underwater surface similarity; Representing historical moments The water flow vector, Indicates the current time The water flow vector; This represents the cosine similarity function for vectors.

2. The rapid crawling control method for a biomimetic adsorption underwater robot according to claim 1, characterized in that, The planned path of the underwater robot is obtained using the following method: Based on a 3D grid map of the underwater surface, the planned path of the underwater robot is obtained using the A-star algorithm.

3. The rapid crawling control method for a biomimetic adsorption underwater robot according to claim 1, characterized in that, The specific method for dividing the planned path into several smaller path segments based on the multi-dimensional environmental data of each grid in the planned path includes: Based on the multi-dimensional environmental data of each grid in the planning path, the segmentation probability of each grid in the planning path is obtained; a coordinate system is constructed with the grid order value as the x-axis and the segmentation probability as the y-axis to obtain the segmentation probability change curve of the planning path. Obtain several maxima in the segmented probability change curve, take the grid corresponding to each maxima as the segment point of the planned path, take the first grid in the planned path as a segment point, and divide the planned path into several small segments through the segment points.

4. The rapid crawling control method for a biomimetic adsorption underwater robot according to claim 3, characterized in that, The segmentation probability of each grid cell in the planned path is obtained using the following method: Each grid cell in the 3D raster map corresponds to a local normal vector, tilt angle, curvature, and underwater depth, with tilt angle, curvature, and underwater depth serving as the data information dimensions for each grid cell; in the planned path, the first... The segmentation probability of each grid cell The calculation method is as follows: in, Indicates the number of dimensions of data information; Indicates the preset reference range; This indicates that in the planned path, starting from the first... The grid to the first The first grid cell The average environmental data across each data information dimension; This indicates that in the planned path, starting from the first... The grid to the first The first grid cell The average environmental data across each data information dimension; Indicates the first path in the planned path The first grid cell Environmental data in multiple dimensions; This represents the function for calculating the standard deviation. This represents the absolute value function.

5. The rapid crawling control method for a biomimetic adsorption underwater robot according to claim 1, characterized in that, The specific method for obtaining the underwater surface similarity between each historical moment and the current moment is as follows: Obtain multi-dimensional environmental data at the current location and construct the current time. underwater surface data vector ,in Indicates the current time The local normal vector at the corresponding position, Indicates the current time The tilt angle at the corresponding position Indicates the current time Curvature at the corresponding position Indicates the current time The underwater depth at the corresponding location; Get historical moments under this short path segment underwater surface data vector ,in Representing historical moments The local normal vector at the corresponding position, Representing historical moments The tilt angle at the corresponding position Representing historical moments Curvature at the corresponding position Representing historical moments The underwater depth at the corresponding location; Historical moment With the current moment underwater surface similarity The calculation method is as follows: in, Representing historical moments underwater surface data vectors, Indicates the current time underwater surface data vector; This indicates the calculation of the Euclidean norm of a vector; To avoid hyperparameters with a denominator of 0; This represents the normalization function.

6. The rapid crawling control method for a biomimetic adsorption underwater robot according to claim 1, characterized in that, The specific method for obtaining the water flow prediction duration at the current moment includes: in, Indicates the current time Water flow prediction duration Indicates the current time Historical water flow stability and These represent the minimum and maximum prediction durations in the DWA algorithm, respectively.

7. The rapid crawling control method for a biomimetic adsorption underwater robot according to claim 1, characterized in that, The specific methods for constructing the robot crawling prediction model are as follows: The input layer is constructed using all moments of the underwater robot up to the current moment, including the water flow vector at each moment, the underwater depth of the robot at the corresponding position at each moment, the robot's crawling direction and the attitude vector composed of the robot's attitude information, as well as several motor powers. The hidden layers of a multilayer perceptron are constructed and activation functions are set; the output layer consists of two neurons, corresponding to linear velocity and angular velocity respectively; through construction and training, the trained multilayer perceptron is finally obtained and used as a robot crawling prediction model.

8. The rapid crawling control method for a biomimetic adsorption underwater robot according to claim 1, characterized in that, Based on the water flow prediction vector, the planned path, and multi-dimensional environmental data at each location along the planned path, a robot crawling prediction model is used to obtain the maximum achievable speed at each moment within the water flow prediction time. Combining this with the speed space, several initial paths are selected for the current moment. The specific methods include: Linear velocity is sampled at a certain sampling rate in the velocity space. With angular velocity Data is collected separately to obtain several linear velocities and several angular velocities. Any linear velocity and any angular velocity are combined to obtain several sets of velocities, and each set of velocities corresponds to a segment of the predicted path. For any given moment within the current flow prediction time, based on the flow prediction vector at that moment, for any segment of the prediction path, there is a corresponding set of linear and angular velocities. The position at that moment within the predicted path segment is obtained, along with the underwater depth and the robot's crawling direction at that position. Simultaneously, based on the local normal vector of the grid at that position, the robot's attitude vector is obtained and input into the robot crawling prediction model. The motor power is set to its maximum value, and the maximum achievable linear velocity and maximum achievable angular velocity at that moment within the predicted path segment are output. The maximum achievable linear velocity and maximum achievable angular velocity at each moment within the flow prediction time within the predicted path segment are also obtained. Under this predicted path segment, if, within a certain time period of the predicted water flow duration, the maximum achievable linear velocity is less than the linear velocity corresponding to this predicted path segment, or the maximum achievable angular velocity is less than the angular velocity corresponding to this predicted path segment, then this predicted path segment is deleted. The above judgment is performed on the predicted paths corresponding to each group of velocities in the velocity space, and the remaining predicted paths are used as several initial paths.

9. The rapid crawling control method for a biomimetic adsorption underwater robot according to claim 1, characterized in that, The specific method for obtaining the optimal path at the current moment from several initial paths using the evaluation function of the DWA algorithm includes: By using distance evaluation coefficient, azimuth evaluation coefficient, and velocity evaluation coefficient, and based on each initial path and a set of linear and angular velocities in the corresponding velocity space, a score is obtained for each initial path, and the initial path with the highest score is taken as the optimal path.

Images