Sneaking and crawling integrated navigation method and system for amphibious robot

Abstract

The invention discloses a sneaking and crawling integrated navigation method and system for an amphibious robot, and relates to the technical field of robot navigation and control, and the method comprises the steps: extracting environment features through a deep learning model based on preprocessed environment information data, and carrying out the feature fusion through employing a Bayesian network, and generating a comprehensive feature vector; based on the fused comprehensive feature vector, establishing a state transition equation by using a Lie group Lie algebra method, and predicting the position and posture change of the amphibious robot; and dynamically adjusting navigation parameters of the amphibious robot through an AI algorithm based on the position and posture change of the amphibious robot. According to the invention, a Bayesian network is adopted to fuse various environment features, such as angular velocity features, acceleration features, depth features and the like, extracted from various sensors, and a comprehensive feature vector is generated. Through the step, effective integration of multi-source information is realized, and the understanding ability of the system to a complex environment is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of robot navigation and control technology, in particular to a stealth and crawling combined navigation method and system for an amphibious robot. Background Art

[0002] As an intelligent device that can move freely underwater and on land, amphibious robots have attracted widespread attention in recent years. Their applications range from ocean exploration, environmental monitoring to military reconnaissance and other fields, showing great potential. With the development of technology, the advancement of sensor technology and machine learning algorithms has provided strong support for the navigation system of amphibious robots. Traditional amphibious robot navigation methods mainly rely on a single type of sensor, such as sonar or visual sensors, for navigation in different environments. However, these early methods have certain limitations. For example, in complex and changing environments, a single type of sensor is difficult to provide enough information to ensure high-precision positioning and attitude estimation. Therefore, researchers began to explore the use of a combination of multiple sensors to improve the performance of the navigation system. By fusing data from different types of sensors, the shortcomings of a single sensor can be effectively compensated and the robustness and adaptability of the system can be improved.

[0003] Although the existing multi-sensor fusion technology has improved the navigation ability of amphibious robots to a certain extent, there are still some significant problems. First, due to the lack of effective feature extraction and fusion mechanism, traditional methods often cannot make full use of the rich information provided by various sensors, resulting in inaccurate navigation decisions. Secondly, in a dynamically changing environment, how to adjust the navigation parameters in real time to cope with environmental changes remains a challenge. Most current methods use fixed models or strategies for navigation control, which limits the autonomy and flexibility of robots in complex and unknown environments. In response to the above problems, the present invention proposes a comprehensive feature extraction and fusion method based on the combination of deep learning and Bayesian networks, and uses the Lie group Lie algebra method to establish the state transfer equation, and dynamically adjusts the navigation parameters through AI algorithms, so as to achieve more accurate position and attitude estimation, as well as higher navigation efficiency. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a stealth and crawling combined navigation method for an amphibious robot to solve the problems in the prior art of insufficient feature extraction and fusion and inability to dynamically adjust navigation parameters in real time.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a stealth and crawling combined navigation method for an amphibious robot, which includes configuring a variety of sensors, including IMU, DVL, sonar, lidar, visual sensor and odometer, pre-calibrating all sensors, and collecting and preprocessing environmental information data; based on the preprocessed environmental information data, extracting environmental features through a deep learning model, and using a Bayesian network to perform feature fusion to generate a comprehensive feature vector, wherein the environmental features include angular velocity features, acceleration features, depth features, echo intensity features, point cloud features, RGB image features, wheel speed features and direction change features; based on the fused comprehensive feature vector, using the Lie group Lie algebra method to establish a state transfer equation to predict the position and posture changes of the amphibious robot; based on the position and posture changes of the amphibious robot, dynamically adjusting the navigation parameters of the amphibious robot through an AI algorithm.

[0008] As a preferred solution of the stealth and crawling combined navigation method of an amphibious robot described in the present invention, wherein: the configuration of multiple sensors, including IMU, DVL, sonar, lidar, visual sensor and odometer, pre-calibrates all sensors, and the specific steps are as follows:

[0009] Pre-calibrate the IMU through static calibration, dynamic calibration and temperature compensation;

[0010] Pre-calibrate DVL through static calibration, dynamic calibration and depth calibration;

[0011] Pre-calibrate the sonar through distance calibration, angle calibration and echo intensity calibration;

[0012] Pre-calibrate the LiDAR through point cloud consistency calibration, angle calibration, and distance calibration;

[0013] The odometer is pre-calibrated through wheel speed calibration, direction calibration and displacement calibration.

[0014] As a preferred solution of the stealth and crawling combined navigation method of an amphibious robot described in the present invention, wherein: the environmental information data includes depth information, speed information, distance information, RGB image information, temperature value and pressure value;

[0015] The preprocessing includes denoising, anomaly detection and processing, calibration and compensation, data interpolation and completion, coordinate transformation and format unification.

[0016] As a preferred solution of the stealth and crawling combined navigation method of an amphibious robot described in the present invention, wherein: based on the pre-processed environmental information data, environmental features are extracted through a deep learning model, and the specific steps are as follows:

[0017] Extract the angular velocity features and acceleration features of the IMU through LSTM;

[0018] Extract the depth features of the DVL through DNN;

[0019] Extract the echo intensity features of the sonar through CNN;

[0020] Extract the point cloud features of the lidar through PointNet;

[0021] Extract the RGB image features of the vision sensor through VGG;

[0022] Extract the wheel speed features and direction change features of the odometer through GRU.

[0023] As a preferred solution of the stealth crawling combined navigation method for an amphibious robot described in the present invention, wherein: the Bayesian network is used for feature fusion to generate a comprehensive feature vector, and the environmental features include angular velocity features, acceleration features, depth features, echo intensity features, point cloud features, RGB image features, wheel speed features and direction change features. The specific steps are as follows.

[0024] Define each environmental feature as an independent node, and establish a dependency relationship by analyzing the conditional dependence and correlation between nodes through PGM;

[0025] Based on the independent nodes and the dependency relationship, establish a Bayesian network;

[0026] Based on the angular velocity features, acceleration features, depth features, echo intensity features, point cloud features, RGB image features, wheel speed features and direction change features, perform feature fusion through the Bayesian network and generate a comprehensive feature vector.

[0027] As a preferred solution of the stealth crawling combined navigation method for an amphibious robot described in the present invention, wherein: based on the fused comprehensive feature vector, use the Lie group and Lie algebra method to establish a state transition equation to predict the position and attitude changes of the amphibious robot. The specific steps are as follows.

[0028] Based on the fused comprehensive feature vector, use the Lie group and Lie algebra method to establish a state transition equation. Represent the position and attitude changes of the amphibious robot as elements on the Lie group, and use the corresponding Lie algebra to describe the small increments of the position and attitude changes of the amphibious robot. Then, convert the small increments back to the Lie group through the exponential map, and predict the position and attitude changes of the amphibious robot through iterative calculation and update within continuous time.

[0029] As a preferred solution of the submersible and crawling combined navigation method for an amphibious robot according to the present invention, wherein: based on the position and attitude changes of the amphibious robot, the navigation parameters of the amphibious robot are dynamically adjusted through an AI algorithm. The specific steps are as follows:

[0030] Based on the position and attitude changes of the amphibious robot, an AI model is trained using reinforcement learning and deep learning;

[0031] Based on the fused comprehensive feature vector, a state space is defined to identify the states of the environment and itself;

[0032] Based on the hardware configuration of the amphibious robot, an action space is defined to identify executable actions;

[0033] Based on the historical task performance, a reward function is set to identify the evaluation criteria;

[0034] The AI model receives the current state information and dynamically adjusts the navigation parameters of the amphibious robot;

[0035] The navigation parameters of the amphibious robot include speed control, path control, attitude adjustment, and obstacle avoidance strategies.

[0036] In a second aspect, the present invention provides a submersible and crawling combined navigation system for an amphibious robot, including a data acquisition and preprocessing module, a feature extraction and fusion module, a motion state update module, and an intelligent navigation and decision-making module; the data acquisition and preprocessing module is used to configure a variety of sensors, including IMU, DVL, sonar, lidar, visual sensors, and odometers, pre-calibrate all sensors, and collect and preprocess environmental information data; the feature extraction and fusion module is used to extract environmental features through a deep learning model based on the preprocessed environmental information data, and perform feature fusion using a Bayesian network to generate a comprehensive feature vector. The environmental features include angular velocity features, acceleration features, depth features, echo intensity features, point cloud features, RGB image features, wheel speed features, and direction change features; the motion state update module is used to establish a state transition equation using Lie group and Lie algebra methods based on the fused comprehensive feature vector to predict the position and attitude changes of the amphibious robot; the intelligent navigation and decision-making module is used to dynamically adjust the navigation parameters of the amphibious robot through an AI algorithm based on the position and attitude changes of the robot.

[0037] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and wherein: when the computer program is executed by the processor, any step of the submersible and crawling combined navigation method for an amphibious robot as described in the first aspect of the present invention is implemented.

[0038] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the stealth and crawling combined navigation method of an amphibious robot as described in the first aspect of the present invention.

[0039] The beneficial effects of the present invention are as follows: by using a Bayesian network to fuse various environmental features such as angular velocity features, acceleration features, depth features, etc. extracted from various sensors, a comprehensive feature vector is generated. This step realizes the effective integration of multi-source information and enhances the system's ability to understand complex environments. Then, based on the fused comprehensive feature vector, the state transfer equation is established using the Lie group Lie algebra method to predict the position and posture changes of the amphibious robot. The advantage of this method is that it can accurately describe the robot's tiny movements, and convert these tiny increments back to actual position and posture changes through exponential mapping, greatly improving the positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0041] Figure 1 This is a flow chart of the stealth and crawling combined navigation method of the amphibious robot in Example 1.

[0042] Figure 2 This is a schematic diagram of the stealth and crawling combined navigation system of the amphibious robot in Example 1. DETAILED DESCRIPTION

[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0044] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0045] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0046] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a stealth and crawling combined navigation method for an amphibious robot, comprising the following steps:

[0047] S1: Configure multiple sensors, including IMU, DVL, sonar, lidar, visual sensor, and odometer, and pre-calibrate all sensors.

[0048] S1.1: Pre-calibrate the IMU through static calibration, dynamic calibration and temperature compensation.

[0049] Furthermore, static calibration is mainly performed when the device is stationary. This process aims to eliminate the zero bias error of the sensor, that is, the phenomenon that the sensor output value is not zero when there is no external acceleration or angular velocity. Static calibration usually requires placing the IMU in a stable position and recording data over a period of time. Through statistical analysis of these data, the sensor bias can be determined and corrected in subsequent data processing. In contrast to static calibration, dynamic calibration is performed when the IMU is in motion. Its purpose is to obtain the response characteristics of the sensor under different motion states, so as to correct the proportional factor error and other nonlinear errors of the sensor. Dynamic calibration usually involves fixing the IMU on a device with a known trajectory, such as a rotating platform or a moving vehicle, and adjusting the output of the sensor by comparing the difference between the actual motion parameters and the IMU measurement results. Since temperature changes will affect the performance of the IMU sensor and cause its output to change, temperature compensation is required to reduce this effect. Temperature compensation can be achieved by pre-establishing a temperature-error model that describes the change of sensor output with temperature. The usual practice is to test the IMU under different temperature conditions, record the changes in the sensor output, and adjust the sensor readings accordingly to maintain high accuracy throughout the entire operating temperature range.

[0050] S1.2: Pre-calibrate the DVL through static calibration, dynamic calibration and depth calibration.

[0051] Furthermore, the main purpose of static calibration is to eliminate the zero-offset error of the sensor. This step is usually performed when the device is completely stationary. By recording the output value of the sensor without the influence of water flow, its inherent bias is determined. The specific operation includes installing the DVL on a stable and known stationary platform and collecting data for a period of time. Analyze this data to determine the bias value and compensate for this bias in subsequent processing. Dynamic calibration is to correct the response characteristics of the DVL under actual motion conditions. Since the DVL relies on detecting the change in the acoustic wave frequency reflected from the bottom of the water or other targets to calculate the speed, its accuracy is affected by various factors, such as the change in the acoustic wave propagation speed, the water flow speed, etc. Dynamic calibration usually needs to be carried out in a controlled environment, such as using a moving platform or carrier, operating at a known speed, and comparing the measurement results of the DVL with the actual speed. Adjust the parameter settings of the DVL according to these differences to improve its accuracy in the real environment. Depth calibration specifically optimizes the performance of the DVL at different depths. Because the speed of sound waves propagating in water varies with water depth, the measurement accuracy of the DVL also changes with depth. Depth calibration usually involves testing under different water depth conditions. By comparing the measurement values of the DVL with reference values (such as the standard values provided by other high-precision instruments), a calibration model is established. This model can be used to adjust the measurement results of the DVL at different depths to ensure that it maintains consistent high accuracy within the entire working depth range.

[0052] S1.3: Pre-calibrate the sonar through distance calibration, angle calibration, and echo intensity calibration.

[0053] Furthermore, distance calibration means ensuring that the sonar can accurately measure the distance to the target. Specifically, set targets with known distances, select one or more reference targets with clear distances, use the sonar device to measure these targets multiple times, and record the results of each measurement. Compare the actual distance with the measurement results, calculate the average error, and adjust the internal parameters of the sonar according to this error to compensate for the measurement deviation. Repeat the above steps until the error between the measured value and the actual value reaches an acceptable range. The purpose of angle calibration is to ensure that the sonar can correctly identify the direction of the target. Set multiple reference points in different directions and ensure that their positional relationship with the sonar is known. Scan these points from different azimuth angles and elevation angles and record the angle readings of each point. Compare the difference between the actual angle and the measured angle, and reduce this deviation by adjusting the mechanical or electronic components of the sonar. Measure the previously used reference points again to check whether the calibration has successfully reduced the angle error. Echo intensity calibration is to optimize the quality of the sonar received signal, that is, the accuracy of the echo intensity. Create a stable test environment containing materials with known reflection characteristics as targets, collect echo intensity data under ideal conditions as the basis for subsequent calibration. Based on the baseline data, appropriately adjust the gain and sensitivity settings of the sonar to match the expected echo intensity level. Further test and fine-tune the settings by changing the distance or reflection characteristics of the target until the sonar can provide consistent and reliable echo intensity readings under various conditions.

[0054] S1.4: Pre-calibrate the lidar through point cloud consistency calibration, angle calibration, and distance calibration.

[0055] Furthermore, point cloud consistency calibration aims to ensure the consistency and accuracy of the point cloud data generated by lidar between different scanning cycles. First, select a static target with a known geometry, such as a plane or a cube, and use the lidar to scan it multiple times. Then, analyze the point cloud data in these scan results to find any inconsistencies or deviations. Adjust the internal algorithm parameters of the lidar according to the analysis results to minimize these errors and ensure that the point cloud data obtained from each scan in subsequent operations can accurately reflect the actual environment. Angle calibration is to ensure that the scanning angles of the lidar in the horizontal and vertical directions are accurate. This process involves mounting the lidar on a platform that can precisely control the rotation angle and then scanning a series of targets at known positions. By comparing the differences between the actual target positions and the angle data recorded by the lidar, the possible angle deviations can be identified. Based on this information, adjust the angle sensor of the lidar or the relevant mechanical components until the measured angle exactly matches the true angle. This step is crucial for ensuring that the lidar can correctly map the surrounding environment. Distance calibration focuses on ensuring that the lidar can accurately measure the distance to an object. First, set a target with a uniform reflectivity at a known distance, and then use the lidar to perform multiple distance measurements on this target. Collect and analyze these measurement data, calculate the mean value and standard deviation, and use them to evaluate the consistency and accuracy of the measurements. If systematic deviations are found, correct this deviation by adjusting the ranging module of the lidar so that the lidar can provide high-precision distance measurements under various conditions. This calibration process is particularly important for improving the reliability and performance of the navigation system of the amphibious robot.

[0056] S1.5: Pre-calibrate the odometer through wheel speed calibration, direction calibration, and displacement calibration.

[0057] Wheel speed calibration is to ensure that the odometer can accurately record the rotational speed of the wheels, so as to accurately calculate the driving distance. First, run the robot on a flat test track with a known length and record the odometer data. By comparing the difference between the actual driving distance and the data recorded by the odometer, it can be determined whether there is an error in the wheel speed sensor. According to these errors, adjust the parameters of the sensor or the mechanical device to make the odometer output closer to the true value. This process helps to improve the accuracy of the odometer at different speeds. Direction calibration aims to ensure that the odometer can correctly identify and record the steering angle of the robot, which is crucial for the accuracy of the navigation system. This process usually involves placing the robot on a platform where the steering angle can be controlled, or marking a series of direction change points on the ground that need to be followed. Then, drive the robot to travel along the predetermined direction change path and record the odometer direction data. Compare the actual direction change with the data recorded by the odometer, find any deviations and adjust the sensor settings or algorithm parameters accordingly to ensure the accuracy of the direction data in subsequent operations. Displacement calibration focuses on verifying and adjusting the overall position perception ability of the odometer, that is, whether the actual displacement of the robot from the starting point to the end point is consistent with the odometer record. Select a closed path, such as a standard-sized rectangle or circular route, and let the robot travel along this path in a complete circle and return to the starting point. During this process, record all relevant data of the odometer, including the driving distance and direction changes, etc. By analyzing the matching degree of these data with the actual path, any cumulative errors can be found and measures can be taken for correction, such as adjusting the initial setting of the odometer or improving the data processing algorithm to reduce the drift error during long-term use and ensure the accuracy of displacement measurement.

[0058] S1.6: Collect environmental information data and perform preprocessing.

[0059] Furthermore, the environmental information data includes depth information, speed information, distance information, RGB image information, temperature value, and pressure value.

[0060] Even further, the preprocessing includes denoising, anomaly detection and processing, calibration and compensation, data interpolation and completion, coordinate transformation, and format unification.

[0061] Specifically, denoising is to eliminate the random noise in the sensor data, and these noises may come from environmental interference or the limitations of the sensor itself. This process usually uses filtering techniques to smooth the data stream and remove those high-frequency noise components that significantly deviate from the normal value. By selecting appropriate filtering parameters, the influence of noise can be minimized while retaining the useful signal, thereby improving the accuracy of subsequent data analysis;

[0062] Anomaly detection and handling aims to identify and correct outliers or anomalies in a dataset. First, a reasonable threshold range is defined, and data points outside this range are considered anomalies. Then, statistical methods or machine learning algorithms are used to automatically detect these outliers. For the detected anomalies, different handling methods can be adopted according to specific situations, such as directly deleting them, replacing them with the average value of adjacent points, or using a prediction model to estimate more reasonable values;

[0063] Calibration and compensation are to correct systematic errors caused by hardware limitations or environmental factors. For example, temperature changes may cause changes in sensor readings, and temperature compensation is required at this time, while measurement deviations between different sensors need to be solved through calibration. Specific operations include obtaining reference data under known conditions and adjusting the sensor output accordingly to ensure that accurate and reliable measurement results can be provided throughout the entire working range;

[0064] Data interpolation and completion are used to fill in missing data points or repair incomplete data sequences. When data at certain time points or positions fails to be successfully collected, interpolation methods can be used to estimate the missing values based on the surrounding data points;

[0065] Coordinate transformation refers to converting data from different sensors into the same coordinate system for comprehensive analysis and fusion. Since each sensor may have different installation positions and orientations, the data they collect is usually in their respective local coordinate systems. To achieve effective information integration, appropriate mathematical transformations must be applied based on the relative relationships between sensors and their motion states to unify all data into the global coordinate system;

[0066] Format unification is to ensure that data from different types of sensors can be processed and analyzed within the same framework. The data formats generated by different sensors may vary significantly, such as text files, binary streams, or custom formats, etc. Therefore, before further processing, all data needs to be converted into a common representation form, such as standardized timestamps, unit systems, and data structures, etc.

[0067] S2: Based on the preprocessed environmental information data, extract environmental features through a deep learning model, and use a Bayesian network to perform feature fusion to generate a comprehensive feature vector.

[0068] S2.1: Extract the angular velocity features and acceleration features of the IMU through LSTM.

[0069] Furthermore, the angular velocity features and acceleration features of the IMU are extracted by LSTM. This mainly utilizes the characteristic that the LSTM network can effectively process time series data. The preprocessed IMU data is input into the LSTM model, and the model automatically identifies and extracts useful angular velocity and acceleration features by learning the time-dependent relationships in the data. These features can reflect the key information of the device's motion state and provide support for subsequent action recognition or behavior analysis. In this process, there is no need to manually design feature extraction rules, and LSTM can autonomously learn the most representative feature expressions based on the input data.

[0070] S2.2: Extract the depth features of the DVL through DNN.

[0071] Furthermore, the depth features of the DVL are extracted by DNN. This mainly utilizes the powerful non-linear mapping ability of DNN to automatically learn and extract high-level abstract information from the original depth data obtained by the DVL, which can effectively represent environmental features. This method avoids the complexity and limitations of manually designing features, making the extracted depth features more accurate and representative, and thus providing reliable data support for the precise positioning and navigation of the amphibious robot.

[0072] S2.3: Extract the echo intensity features of the sonar through CNN.

[0073] Furthermore, the echo intensity features of the sonar are extracted by CNN. This mainly utilizes the advantages of CNN in processing grid data such as images to automatically learn complex patterns and features from the sonar echo intensity data. This method can effectively identify and extract feature information related to environmental structure and target characteristics, providing strong support for subsequent target detection, classification, and environmental understanding, and at the same time improving the accuracy and efficiency of sonar data processing.

[0074] S2.4: Extract the point cloud features of the lidar through PointNet.

[0075] Furthermore, the point cloud features of the lidar are extracted by PointNet. This mainly utilizes the characteristic that the PointNet network can directly process unordered point cloud data to automatically learn and extract representative geometric and spatial features from the point cloud collected by the lidar. This method not only overcomes the limitations of traditional methods that require voxelization or generating image representations when processing point cloud data, but also can efficiently capture the key information in the point cloud, providing strong feature support for applications such as 3D object recognition and scene understanding.

[0076] S2.5: Extract the RGB image features of the visual sensor through VGG.

[0077] Furthermore, the RGB image features of the visual sensor are extracted by VGG. This is mainly to utilize the powerful representation ability of the deep architecture of the VGG network to automatically learn rich hierarchical features from the RGB images. These features cover various information from simple edges to complex object structures. This method can effectively capture the key visual elements in the images and provide accurate feature support for subsequent image classification, object recognition, and scene analysis.

[0078] S2.6: Extract the wheel speed features and direction change features of the odometer through GRU.

[0079] Extract the wheel speed features and direction change features of the odometer through GRU. This is mainly to utilize the ability of GRU to process sequential data to automatically learn the inherent temporal dependence and dynamic characteristics from the time series of wheel speed and direction change data. This method can capture the subtle changes during the movement of the device, provide strong feature support for accurate position tracking and path prediction, and at the same time improve the understanding and prediction ability of complex motion patterns.

[0080] S2.7: Use a Bayesian network for feature fusion to generate a comprehensive feature vector.

[0081] Define each environmental feature as an independent node, and establish a dependency relationship by analyzing the conditional dependence and correlation between nodes through PGM;

[0082] Based on the independent nodes and dependency relationships, establish a Bayesian network;

[0083] Based on the angular velocity features, acceleration features, depth features, echo intensity features, point cloud features, RGB image features, wheel speed features, and direction change features, perform feature fusion through a Bayesian network and generate a comprehensive feature vector. The expression is:

[0084]

[0085] where N represents the number of feature types, i represents the index variable of the feature type, w i represents the importance weight of the i-th type of feature, P(F|X i ) represents the probability of the comprehensive feature vector F given the i-th type of feature, F represents the comprehensive feature vector, and X i represents the specific feature vector of the i-th type of feature.

[0086] S3: Based on the fused comprehensive feature vector, use the Lie group and Lie algebra method to establish a state transition equation to predict the position and attitude changes of the amphibious robot.

[0087] Furthermore, based on the fused comprehensive feature vector, a state transition equation is established using the Lie group and Lie algebra method. By representing the position and attitude changes of the amphibious robot as elements on the Lie group and using the corresponding Lie algebra to describe the small increments of the position and attitude changes of the amphibious robot, and then converting the small increments back to the Lie group through the exponential map, the position and attitude changes of the amphibious robot are predicted through iterative calculations and updates within continuous time. The specific expression is:

[0088]

[0089] where k represents the current moment, k + 1 represents the next time step immediately following, x k represents the position and attitude of the amphibious robot at time step k, and x k+1 represents the state vector at the next time step k + 1 predicted based on the current state x k ω k represents the angular velocity vector at the current moment k, Δt represents the time interval from time k to k + 1, v k represents the linear velocity vector at the current moment k, 0 T represents the zero vector, and 0 represents its scalar form.

[0090] Specifically, by calculating the exponential map result of x k+1 relative to x k the position and attitude changes of the amphibious robot at the next time step are directly predicted.

[0091] S4: Based on the position and attitude changes of the amphibious robot, the navigation parameters of the amphibious robot are dynamically adjusted through an AI algorithm.

[0092] S4.1: Based on the position and attitude changes of the amphibious robot, an AI model is trained using reinforcement learning and deep learning

[0093] Furthermore, training an AI model using reinforcement learning and deep learning based on the position and attitude changes of the amphibious robot means that by combining the powerful feature extraction ability of the deep neural network and the decision optimization mechanism of reinforcement learning, the AI model can learn the state representation in a complex environment from a large amount of sensor data, and automatically explore the optimal control strategy according to the defined state space, action space, and reward function. This method allows the robot to autonomously navigate in different media, adjust its position and attitude in real time to adapt to the dynamically changing environment and complete specific tasks. In this way, the AI model can not only understand the current environmental information but also predict future state changes, thus making the best action selection to ensure efficient and safe operation.

[0094] S4.2: Based on the fused comprehensive feature vector, define the state space and identify the states of the environment and itself.

[0095] Furthermore, based on the fused comprehensive feature vector, defining the state space and identifying the states of the environment and itself means integrating data from multiple sensors to form a comprehensive feature representation for accurately describing the current position, attitude of the robot, and the state of its surrounding environment. This approach enables the AI system to understand and process complex and changing information within a unified framework, thereby providing an accurate basis for decision-making and control, ensuring that the robot can effectively execute tasks and adapt to various challenges in a dynamic environment.

[0096] S4.3: Based on the hardware configuration of the amphibious robot, define the action space and identify the executable actions.

[0097] Furthermore, based on its own hardware configuration, defining the action space and identifying the executable actions means determining the set of all possible actions that the robot can perform according to its specific mechanical structure and driving capabilities. This approach ensures that all actions considered during the planning and control processes are actually achievable by the robot, enabling the robot to effectively select and execute the optimal actions in a given task environment to achieve the goals and meet various operation requirements.

[0098] S4.4: Based on the historical task performance, set the reward function and identify the evaluation criteria.

[0099] Furthermore, based on the task requirements and environmental feedback, setting the reward function and identifying the evaluation criteria means designing an incentive mechanism to evaluate the quality of the robot's behavior according to the task goals that the robot needs to complete and the real-time information from the environment. Through positive rewards and negative punishments, guide the robot to learn the optimal strategy, ensuring that it can make the best decisions when executing tasks and adapt to changing environmental conditions. This approach enables the robot to dynamically adjust its behavior pattern to efficiently and safely achieve the task goals.

[0100] S4.5: The AI model receives the current state information and dynamically adjusts the navigation parameters of the amphibious robot.

[0101] Furthermore, the AI model receiving the current state information and dynamically adjusting the navigation parameters of the amphibious robot means that by real-time analyzing the data from sensors, the AI model can intelligently evaluate the current situation and automatically optimize the robot's navigation settings, such as speed, direction, and path planning. This approach enables the robot to flexibly respond to various challenges in a complex and changing environment, ensuring efficient and safe arrival at the destination or completion of the specified task. Through continuous learning and adaptation, the AI model continuously improves the decision-making quality and enhances the robot's autonomous operation ability in different scenarios.

[0102] The present embodiment also provides a stealth and crawling integrated navigation system for an amphibious robot, comprising: a data acquisition and preprocessing module, a feature extraction and fusion module, a motion state update module and an intelligent navigation and decision module; the data acquisition and preprocessing module is used to configure a variety of sensors, including IMU, DVL, sonar, lidar, visual sensor and odometer, pre-calibrate all sensors, and collect environmental information data and perform preprocessing; the feature extraction and fusion module is used to extract environmental features through a deep learning model based on the preprocessed environmental information data, and use a Bayesian network to perform feature fusion to generate a comprehensive feature vector, wherein the environmental features include angular velocity features, acceleration features, depth features, echo intensity features, point cloud features, RGB image features, wheel speed features and direction change features; the motion state update module is used to establish a state transfer equation based on the fused comprehensive feature vector using the Lie group and Lie algebra method to predict the position and posture changes of the amphibious robot; the intelligent navigation and decision module is used to dynamically adjust the navigation parameters of the amphibious robot through an AI algorithm based on the position and posture changes of the robot.

[0103] This embodiment also provides a computer device, which is suitable for the stealth and crawling combined navigation method of an amphibious robot, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the stealth and crawling combined navigation method of an amphibious robot as proposed in the above embodiment.

[0104] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0105] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the combined navigation method for amphibious robots' stealth crawling proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM for short), electrically erasable programmable read-only memory (EEPROM for short), erasable programmable read-only memory (EPROM for short), programmable read-only memory (PROM for short), read-only memory (ROM for short), magnetic memory, flash memory, magnetic disk or optical disc.

[0106] In summary, the present invention: fuses various environmental features such as angular velocity features, acceleration features, and depth features extracted from multiple sensors by using a Bayesian network to generate a comprehensive feature vector. This step realizes the effective integration of multi-source information and enhances the system's understanding ability of complex environments. Then, based on the fused comprehensive feature vector, a state transition equation is established using the Lie group and Lie algebra method to predict the position and attitude changes of the amphibious robot. The advantage of this method is that it can accurately describe the tiny movements of the robot and convert these tiny increments back to actual position and attitude changes through exponential mapping, greatly improving the positioning accuracy.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A stealth and crawling combined navigation method for an amphibious robot, characterized in that: include, Configure multiple sensors, including IMU, DVL, sonar, lidar, visual sensor and odometer, pre-calibrate all sensors, collect environmental information data and pre-process them; Based on the preprocessed environmental information data, the environmental features are extracted through a deep learning model, and the Bayesian network is used to fuse the features to generate a comprehensive feature vector. The environmental features include angular velocity features, acceleration features, depth features, echo intensity features, point cloud features, RGB image features, wheel speed features, and direction change features. Based on the fused comprehensive feature vector, the state transfer equation is established using the Lie group and Lie algebra method to predict the position and posture changes of the amphibious robot. Based on the position and posture changes of the amphibious robot, the navigation parameters of the amphibious robot are dynamically adjusted through AI algorithms.

2. The stealth and crawling combined navigation method of an amphibious robot as claimed in claim 1, characterized in that: The configuration includes multiple sensors, including IMU, DVL, sonar, lidar, visual sensor and odometer, and pre-calibration of all sensors. The specific steps are as follows: Pre-calibrate the IMU through static calibration, dynamic calibration and temperature compensation; Pre-calibrate DVL through static calibration, dynamic calibration and depth calibration; Pre-calibrate the sonar through distance calibration, angle calibration and echo intensity calibration; Pre-calibrate the LiDAR through point cloud consistency calibration, angle calibration, and distance calibration; The odometer is pre-calibrated through wheel speed calibration, direction calibration and displacement calibration.

3. The stealth and crawling combined navigation method of an amphibious robot as claimed in claim 2, characterized in that: The environmental information data includes depth information, speed information, distance information, RGB image information, temperature value and pressure value; The preprocessing includes denoising, anomaly detection and processing, calibration and compensation, data interpolation and completion, coordinate transformation and format unification.

4. The stealth and crawling combined navigation method of an amphibious robot as claimed in claim 3, characterized in that: The environmental information data after preprocessing is used to extract environmental features through a deep learning model. The specific steps are as follows: The angular velocity features and acceleration features of IMU are extracted through LSTM; Extract the deep features of DVL through DNN; Extract the echo intensity features of sonar through CNN; Extract the point cloud features of the lidar through PointNet; Extract the RGB image features of the visual sensor through VGG; The wheel speed features and direction change features of the odometer are extracted through GRU.

5. The stealth and crawling combined navigation method of an amphibious robot as claimed in claim 4, characterized in that: Based on the preprocessed environmental information data, the environmental features are extracted through a deep learning model, and the Bayesian network is used to fuse the features to generate a comprehensive feature vector. The specific steps are as follows: Each environmental feature is defined as an independent node, and the dependency relationship is established by analyzing the conditional dependencies and correlations between nodes through PGM; Building a Bayesian network based on independent nodes and dependencies; Based on angular velocity features, acceleration features, depth features, echo intensity features, point cloud features, RGB image features, wheel speed features and direction change features, feature fusion is performed through a Bayesian network to generate a comprehensive feature vector.

6. The stealth and crawling combined navigation method of an amphibious robot as claimed in claim 5, characterized in that: Based on the fused comprehensive feature vector, the state transfer equation is established by using the Lie group Lie algebra method to predict the position and posture changes of the amphibious robot. The specific steps are as follows: Based on the fused comprehensive feature vector, the state transfer equation is established using the Lie group and Lie algebra method. The position and posture changes of the amphibious robot are represented as elements on the Lie group, and the corresponding Lie algebra is used to describe the tiny increments of the position and posture changes of the amphibious robot. The tiny increments are then converted back to the Lie group through exponential mapping. The position and posture changes of the amphibious robot are predicted through iterative calculation and updating in continuous time.

7. The stealth and crawling combined navigation method of an amphibious robot as claimed in claim 6, characterized in that: The specific steps of dynamically adjusting the navigation parameters of the amphibious robot through the AI ​​algorithm based on the position and posture changes of the amphibious robot are as follows: Based on the position and posture changes of the amphibious robot, the AI ​​model is trained using reinforcement learning and deep learning; Based on the fused comprehensive feature vector, the state space is defined to identify the state of the environment and itself; Based on the hardware configuration of the amphibious robot, define the action space and identify executable actions; Based on historical task performance, set reward functions and identify evaluation criteria; The AI ​​model receives the current status information and dynamically adjusts the navigation parameters of the amphibious robot; The navigation parameters of the amphibious robot include speed control, path control, posture adjustment and obstacle avoidance strategy.

8. A stealth and crawling combined navigation system for an amphibious robot, based on a stealth and crawling combined navigation method for an amphibious robot according to any one of claims 1 to 7, characterized in that: Including data acquisition and preprocessing module, feature extraction and fusion module, motion state update module and intelligent navigation and decision module; Data acquisition and preprocessing module, used to configure multiple sensors, including IMU, DVL, sonar, lidar, visual sensor and odometer, pre-calibrate all sensors, and collect and pre-process environmental information data; The feature extraction and fusion module is used to extract environmental features based on the preprocessed environmental information data through a deep learning model, and use a Bayesian network to fuse the features to generate a comprehensive feature vector. The environmental features include angular velocity features, acceleration features, depth features, echo intensity features, point cloud features, RGB image features, wheel speed features, and direction change features. The motion state update module is used to establish the state transfer equation based on the fused comprehensive feature vector using the Lie group Lie algebra method to predict the position and posture changes of the amphibious robot; The intelligent navigation and decision-making module is used to dynamically adjust the navigation parameters of the amphibious robot through AI algorithms based on the robot's position and posture changes.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the stealth and crawling combined navigation method of an amphibious robot as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the stealth and crawling combined navigation method of an amphibious robot as described in any one of claims 1 to 7 are implemented.

Images