An autonomous environment recognition and cross-domain motion mode switching method and device for a triphibian robot

Abstract

The application discloses an autonomous environment recognition and cross-domain motion mode switching method and device of a triphibian robot, wherein the method comprises the following steps: a main controller collects output data of a multi-source sensor, and establishes a real-time data buffer queue; a sliding time window is constructed according to a preset time length based on the multi-source sensor data collected continuously, feature extraction is performed on the data in each sliding time window, and an environment feature vector is constructed; the environment feature vector is input into a pre-trained multi-layer perception environment recognition model, and the current environment in which the robot is located is classified and recognized; prediction probabilities of the robot belonging to three categories of air environment, ground environment and water environment are output, the most possible environment in which the robot is currently located is determined, and environment recognition reliability is further calculated; the main controller judges whether the robot has stably entered a target environment; and a mode switching control module automatically completes corresponding motion mode switching according to the environment recognition result.

Description

Technical Field

[0001] This invention relates to the field of robot autonomous control technology, and in particular to a method and apparatus for autonomous environment recognition and cross-domain motion mode switching of a tri-amphibious robot. Background Technology

[0002] With the rapid development of unmanned systems technology, intelligent robot technology, and artificial intelligence technology, amphibious robots with the ability to operate autonomously across environments are gradually becoming an important research direction in the field of robotics. Compared with traditional robots that can only work in a single environment, amphibious robots can autonomously switch their movement modes between different environments such as the air, ground, and water. They have the characteristics of strong environmental adaptability, wide task execution range, and excellent maneuverability, and have broad application prospects in fields such as disaster relief, environmental monitoring, power line inspection, marine resource exploration, military reconnaissance, and emergency material delivery.

[0003] In recent years, a great deal of research has been conducted both domestically and internationally on amphibious robots. For example, air-ground robots can quickly traverse obstacles using flight mode and achieve efficient ground movement using wheeled mode; air-water robots can complete aerial flight and water surface cruising tasks; some tri-amphibious robots have further integrated flight mechanisms, ground movement mechanisms, and water surface propulsion mechanisms, realizing multiple movement modes such as aerial flight, ground movement, and water surface navigation. However, most current research on amphibious robots mainly focuses on robot mechanism design, power system design, and multi-environment motion performance optimization, while research on how robots can accurately identify their current environment and achieve autonomous movement mode switching is relatively limited.

[0004] Currently, most amphibious robots still rely primarily on manual remote control to switch between different motion modes in practical applications. For example, when a robot descends from the air to the ground, the operator needs to manually switch to ground motion mode based on the situation; similarly, when the robot enters water, it needs to be manually switched to water propulsion mode. This method not only depends on the operator's experience but is also prone to failure in complex environments due to judgment delays, communication latency, or human error, leading to problems such as robot instability, power system conflicts, and task execution failures. This makes it difficult to meet the development needs of autonomous robot operation in complex environments.

[0005] To improve robot autonomy, some recent studies have begun to explore the use of sensors to automatically identify the robot's environment. For example, GPS altitude information can be used to determine if the robot is in flight, contact switches can be used to detect if the robot is touching the ground, pressure or water level sensors can be used to determine if the robot is underwater, or inertial measurement units (IMUs) can be used to collect the robot's motion state information and combined with threshold methods for environmental identification. These methods are simple to implement, computationally inexpensive, and can achieve basic environmental assessment under specific conditions. However, because the robot's motion state in different environments is highly uncertain, relying solely on a single sensor or fixed thresholds for assessment is easily affected by changes in robot posture, acceleration, and external environmental disturbances, resulting in low accuracy in environmental identification and failing to meet the requirements for stable and reliable autonomous mode switching in complex environments.

[0006] Especially in the application of amphibious robots, there are significant dynamic differences between the three environments: air, ground, and water, although some sensor outputs show certain similarities. For example, the vertical impact during a robot's descent and the inertial changes generated during ground acceleration have similar acceleration characteristics; the attitude changes caused by water wave disturbances and the attitude fluctuations caused by wind disturbances during flight may also have similar inertial measurement characteristics. Therefore, relying solely on inertial measurement data is insufficient to accurately distinguish the robot's current environment, while simply adding threshold judgment conditions can easily lead to problems such as increased system complexity, decreased generalization ability, and insufficient environmental adaptability.

[0007] On the other hand, to improve water environment recognition capabilities, some systems employ water level sensors, conductive sensors, or liquid level detection sensors as the basis for environmental detection. When the robot enters the water surface, the water level sensor can quickly detect liquid contact signals, thus achieving water environment recognition. However, in practical applications, when the robot flies in rainy, foggy, high-humidity environments, or areas with splashing waves, water mist, water vapor in the air, and water droplets adhering to the sensor surface can all cause the water level sensor to be falsely triggered, leading the system to mistakenly believe that the robot has entered a water environment. If the control system directly switches modes based on the output of the water level sensor, it may incorrectly switch to water surface movement mode while the robot is still in flight, causing the flight controller to prematurely disengage, or even causing the robot to become unstable and fall. Therefore, how to improve the robustness of the environmental recognition system to sensor false triggering and complex environmental interference has become an important technical problem that urgently needs to be solved in the field of autonomous control of amphibious robots.

[0008] Furthermore, traditional machine learning algorithms such as Support Vector Machines (SVM) and Decision Trees do not adequately utilize the correlations between information from multiple sensor sources and have limited resistance to environmental noise and abnormal data. When the robot is in an environment transition phase or subjected to complex external disturbances, the classification results are prone to frequent jumps, leading to frequent switching of motion modes and affecting the robot's motion stability and system reliability. Therefore, simply improving classification accuracy cannot completely solve the problems existing in the autonomous control of amphibious robots. How to combine multi-source sensor information to achieve stable identification of environmental states and further establish a reliable motion mode switching strategy are key technical issues that need to be addressed in the field of autonomous control of amphibious robots.

[0009] Therefore, in view of the above-mentioned problems in the existing technology, there is an urgent need to propose a method for autonomous environmental recognition and motion mode switching of amphibious robots that can integrate information from multiple sources, has strong environmental adaptability and anti-interference ability, so as to realize the robot's autonomous recognition of three environments: air, ground and water, and stably complete motion mode switching based on the recognition results, thereby improving the robot's autonomous operation capability across environments, motion stability and task execution reliability in complex environments. Summary of the Invention

[0010] This invention provides a method and apparatus for autonomous environment recognition and cross-domain motion mode switching of an amphibious robot, aiming to at least solve one of the technical problems existing in the prior art. The method for autonomous environment recognition and cross-domain motion mode switching of the amphibious robot is applied to an amphibious robot, which includes at least a robot body equipped with a flight drive mechanism, a ground movement mechanism, and a water surface propulsion mechanism, a multi-source sensor module mounted on the robot body, and a main controller. The multi-source sensors include at least an inertial measurement unit and a water level sensor. The method for autonomous environment recognition and cross-domain motion mode switching of the amphibious robot includes the following steps: S100: The main controller synchronously collects the output data of the multi-source sensors according to the preset sampling frequency, and uses a unified timestamp method to synchronize the output data of the multi-source sensors and establish a real-time data cache queue. S200 and the main controller construct a sliding time window based on the continuously collected multi-source sensor data according to a preset time length, and extract features from the data in each sliding time window. In each sliding time window, the output data of the inertial measurement unit and the state of the water level sensor are fused to construct an environmental feature vector that can reflect the robot's current motion environment. The S300 main controller inputs environmental feature vectors into a pre-trained multilayer perceptron environment recognition model to classify and identify the robot's current environment. S400 After the multilayer perceptron environment recognition model completes the classification, it outputs the predicted probability of the robot belonging to three categories: aerial environment, ground environment, and water environment. The main controller determines the most likely environment where the robot is currently located based on the predicted probability corresponding to the three categories, and further calculates the credibility of the environment recognition. The S500 main controller continuously saves the environmental recognition results corresponding to multiple sliding time windows, determines whether the environmental recognition results are consistent within multiple consecutive time windows, and determines whether the robot has stably entered the target environment based on at least one latest environmental recognition credibility assessment result. S600: Once it is determined that the robot has stably entered the target environment, the mode switching control module automatically completes the corresponding motion mode switching based on the environment recognition result. S700, cyclically execute steps S100 to S600, continuously collect robot operating status information, and cyclically execute environment recognition, environment judgment and motion mode switching control according to the above steps.

[0011] Furthermore, this invention also proposes a triphibious robot for executing the autonomous environment recognition and cross-domain movement mode switching method of the triphibious robot. The triphibious robot has the ability to fly in the air, move on the ground, and navigate on the water. The triphibious robot includes: The robot body is equipped with an actuator and control module and a power management module. The actuator and control module includes at least a flight drive mechanism, a ground movement mechanism and a water propulsion mechanism. A multi-source sensor module is mounted on the robot body. The multi-source sensor module includes at least an inertial measurement unit and a water level sensor. The inertial measurement unit (IMU) is installed near the center of the robot body and is used to measure the robot's three-axis acceleration and three-axis angular velocity information in real time. The IMU outputs the linear acceleration and angular velocity of the robot in three directions, namely the x-axis acceleration, y-axis acceleration, z-axis acceleration, and angular velocity around the x-axis, y-axis, and z-axis, respectively. The water level sensor is used to detect whether the robot is in contact with water and is installed on the bottom of the robot. The main controller is used to perform environmental recognition calculations and motion mode decisions. The multi-source sensor module is electrically connected to the main controller. The main controller is equipped with an environmental recognition module and a motion mode switching control module. The environmental recognition module uses a trained environmental classification model to determine the robot's current environment in real time based on inertial measurement data and water level sensor data. The motion mode switching module manages the flight controller, ground controller, and water surface controller in a unified manner based on the output results of the environmental recognition module, and automatically completes the controller switching.

[0012] The beneficial effects of this invention are as follows: This invention proposes a method for autonomous environment recognition and cross-domain motion mode switching for amphibious robots, which solves the technical problems of existing amphibious robots, such as lack of autonomous environment recognition capability among the three motion environments of air, ground and water, reliance on manual intervention for motion mode switching, insufficient reliability of single sensor recognition, and mode switching errors in complex environments. This improves the robot's autonomous operation capability across environments and its operational stability in complex environments. Attached Figure Description

[0013] Figure 1 This is a flowchart of the autonomous environment recognition and cross-domain motion mode switching method of the amphibious robot of the present invention.

[0014] Figure 2 This is a schematic diagram of the logic and information flow of the amphibious robot in an embodiment of the present invention.

[0015] Figure 3 This is a schematic diagram of the robot hardware system in an embodiment of the present invention.

[0016] Figure 4 This is a schematic diagram of the robot control software system in an embodiment of the present invention.

[0017] Figure 5 This is a schematic diagram illustrating how the main controller uses a sliding time window to process continuous data in an embodiment of the present invention.

[0018] Figure 6 This is a schematic diagram of a multilayer perceptron (MLP) as an environmental classification model in an embodiment of the present invention.

[0019] Figure 7 This is the result of the robot's autonomous environment recognition under conditions without environmental interference in this embodiment of the invention.

[0020] Figure 8 This is the result of the robot's autonomous environment recognition after environmental interference was introduced in an embodiment of the present invention. Detailed Implementation

[0021] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with the embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention.

[0022] It should be noted that, unless otherwise specified, when a feature is referred to as "fixed" or "connected" to another feature, it can be directly fixed or connected to the other feature, or indirectly fixed or connected to the other feature. The singular forms "a," "described," and "the" used herein are also intended to include the plural forms, unless the context clearly indicates otherwise. Furthermore, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing particular embodiments only and not for limiting the invention. The term "and / or" as used herein includes any combination of one or more of the associated listed items.

[0023] Reference Figures 1 to 8 This invention provides a method for autonomous environment recognition and cross-domain motion mode switching of a triphibious robot, aiming to solve at least one of the technical problems existing in the prior art. (Refer to...) Figure 1 The autonomous environment recognition and cross-domain motion mode switching method for the amphibious robot is applied to an amphibious robot. The amphibious robot includes at least a robot body equipped with a flight drive mechanism, a ground movement mechanism, and a water surface propulsion mechanism, a multi-source sensor module and a main controller mounted on the robot body, and the multi-source sensors include at least an inertial measurement unit and a water level sensor. The autonomous environment recognition and cross-domain motion mode switching method for the amphibious robot includes the following steps: S100: The main controller synchronously collects the output data of the multi-source sensors according to the preset sampling frequency, and uses a unified timestamp method to synchronize the output data of the multi-source sensors and establish a real-time data cache queue. For step S100, in order to improve the consistency of data between different sensors, the present invention preferably adopts a unified timestamp method to complete the synchronization of multi-source data in the multi-source sensor data acquisition step, and establishes a real-time data cache queue so that the data collected by different sensors can form a unified data sequence in chronological order, ensuring that various features have a consistent data source in the subsequent environmental identification process and improving the accuracy of environmental identification.

[0024] S200 and the main controller construct a sliding time window based on the continuously collected multi-source sensor data according to a preset time length, and extract features from the data in each sliding time window. In each sliding time window, the output data of the inertial measurement unit and the state of the water level sensor are fused to construct an environmental feature vector that can reflect the robot's current motion environment. Specifically, since the robot's motion state typically undergoes continuous changes as it moves from the air to the ground or water surface, or from the water surface back to the air, using a sliding time window can effectively reduce the impact of random noise in single-frame data on the environmental recognition results, making environmental recognition more stable and reliable. Simultaneously, multi-source feature fusion can overcome the problem of single sensors being easily affected by local environmental interference, providing a richer and more reliable data foundation for subsequent environmental classification.

[0025] The purpose of step S200 is to construct a sliding time window and extract environmental features. In a specific embodiment, the sliding window length is set to 0.5 s, and adjacent windows are updated by sliding with a fixed step size to ensure that the robot's environmental state can be continuously updated in real time.

[0026] The environmental features described can characterize the robot's posture changes, motion stability, and inertial change patterns, as well as reflect state information such as whether the robot is in contact with a water environment. This fully utilizes the complementary advantages of multiple source sensors and improves the separability between different motion environments.

[0027] The S300 main controller inputs environmental feature vectors into a pre-trained multi-layer perceptron (MLP) environmental recognition model to classify and identify the robot's current environment. S400: After the Multi-Layer Perceptron (MLP) environment recognition model completes the classification, it outputs the predicted probability of the robot belonging to three categories: aerial environment, ground environment, and water environment. The main controller determines the most likely environment where the robot is currently located based on the predicted probability corresponding to the three categories, and further calculates the credibility of the environment recognition. For step S400, environmental confidence assessment can prevent drastic fluctuations in classification results near the environmental boundary due to sensor noise, posture disturbances, or motion impacts, thereby reducing the problem of frequent switching of motion modes caused by accidental misjudgments. For example, when the robot is in the transition from air to ground, due to the significant changes in vertical velocity and posture, the environmental recognition model may simultaneously output high probabilities for both the air and ground environments. In this case, the environmental confidence is low, and the system will not immediately switch motion modes but will wait for further stabilization of environmental information, thus improving the overall stability of the control system. Similarly, when the robot is flying in rainy or foggy environments or areas with splashing waves, even if the water level sensor is falsely triggered by water droplets, the inertial measurement unit still reflects that the robot is in flight, and the environmental recognition model can still maintain a high probability of the air environment, thereby preventing the robot from mistakenly entering the water surface motion mode and improving the robustness of the environmental recognition system to complex environmental disturbances.

[0028] The S500 main controller continuously saves the environmental recognition results corresponding to multiple sliding time windows, determines whether the environmental recognition results are consistent within multiple consecutive time windows, and determines whether the robot has stably entered the target environment based on at least one latest environmental recognition credibility assessment result. Specifically, regarding the continuous window consistency confirmation and mode switching determination method, compared to the traditional method of directly completing mode switching based on a single recognition result, this invention effectively reduces erroneous switching caused by short-term sensor fluctuations, random noise, and changes in environmental boundaries through a continuous window confirmation mechanism, thereby improving the robot's stability across environments. For example, when a robot enters water from the air, the inertial measurement unit output may change drastically due to the impact upon contact with the water surface, and the water level sensor may also experience short-term vibrations. If mode switching is directly completed based on a single recognition result, it can easily lead to frequent switching by the robot controller. This invention requires that multiple consecutive time windows remain consistent before switching can be performed, effectively avoiding oscillations in the control system and improving the overall operational reliability of the robot.

[0029] S600: Once it is determined that the robot has stably entered the target environment, the mode switching control module automatically completes the corresponding motion mode switching based on the environment recognition result. S700, cyclically execute steps S100 to S600, continuously collect robot operating status information, and cyclically execute environment recognition, environment judgment and motion mode switching control according to the above steps.

[0030] Specifically, the entire recognition and control process constitutes a closed loop for the robot's autonomous operation, enabling the robot to update its motion mode in real time according to environmental changes, and achieve continuous autonomous switching between three motion states: air, ground, and water, thereby improving the robot's ability to operate continuously in complex environments. Because the entire control process adopts an online real-time operation mode, it is applicable to various application scenarios such as robot obstacle crossing, air-to-ground transition, water surface take-off and landing, water surface cruising, and autonomous operation in complex environments.

[0031] The beneficial effects of this invention are as follows: This invention proposes a method for autonomous environment recognition and cross-domain motion mode switching for amphibious robots, which solves the technical problems of existing amphibious robots, such as lack of autonomous environment recognition capability among the three motion environments of air, ground and water, reliance on manual intervention for motion mode switching, insufficient reliability of single sensor recognition, and mode switching errors in complex environments. This improves the robot's autonomous operation capability across environments and its operational stability in complex environments.

[0032] The amphibious robot has three movement capabilities: aerial flight, ground movement, and water navigation. It can automatically switch movement modes according to different environments, enabling the robot to operate autonomously across environments.

[0033] Existing amphibious robots typically integrate flight mechanisms, ground movement mechanisms, and water propulsion mechanisms. When performing cross-environment tasks, they need to switch between appropriate movement modes based on the environment. However, most robots currently rely on manual remote control to switch between flight, ground, and water modes, as the robots themselves cannot autonomously determine their current environment. When performing tasks such as obstacle crossing, air-to-ground transitions, water landings, or water takeoffs, operators must promptly switch control modes based on the robot's movement status. This not only increases the operator's workload but also makes them susceptible to errors in mode switching due to human judgment, communication delays, or complex environmental factors, thus reducing the robot's autonomous operation capabilities.

[0034] On the other hand, some existing studies have attempted to use single sensors such as inertial measurement units (IMUs), pressure sensors, contact sensors, and water level sensors to achieve environmental recognition. However, due to the complex dynamic characteristics of robots in different motion environments, single sensors are easily affected by attitude changes, motion impacts, environmental noise, and external interference, resulting in low accuracy in environmental recognition. For example, the inertial impact generated during the robot's descent can easily produce similar acceleration changes to ground acceleration; the attitude changes caused by wave disturbances during the robot's floating on water may also be similar to the attitude changes caused by wind disturbances during flight. Therefore, relying solely on inertial information is insufficient to accurately distinguish the robot's current environment.

[0035] Furthermore, in practical applications, when robots fly in rainy or foggy weather, high-humidity environments, or areas with splashing waves, water mist, water vapor in the air, and water droplets adhering to the surface of the water level sensor can all cause the water level sensor to trigger falsely, leading the system to mistakenly believe that the robot has entered a water environment. If the robot directly switches motion modes based on the water level sensor output, it may prematurely switch to water propulsion mode while still in flight, causing the flight controller to malfunction, and in severe cases, even leading to robot instability or crashes. Therefore, improving the robustness of robot environmental recognition systems to complex environmental interference and abnormal sensor inputs is a crucial technical problem that urgently needs to be solved in the field of autonomous control of amphibious robots.

[0036] In view of this, the present invention proposes an autonomous environment recognition and motion mode switching method for a triphibious robot. By integrating multi-source information such as inertial measurement unit and water level sensor, and combining time sliding window, feature extraction, environment classification, environment confidence assessment and continuous window consistency confirmation mechanism, the robot can autonomously recognize three environments: air, ground and water. Based on the environment recognition results, it can automatically complete the corresponding motion mode switching and controller switching. While ensuring the accuracy of environment recognition, it can effectively reduce the impact of sensor false triggering in complex environments, and improve the robot's autonomous operation capability, motion stability and control reliability across environments.

[0037] Compared with the prior art, the present invention has at least the following beneficial effects: (1) This invention proposes an autonomous environment recognition and motion mode switching method for amphibious robots, which realizes the robot's autonomous recognition of three typical motion environments: air, ground and water. The motion mode can be automatically switched without manual remote control, thus improving the robot's autonomous operation capability.

[0038] (2) This invention integrates information from multiple sources such as inertial measurement units and water level sensors, making full use of the complementary advantages between different sensors. Compared with single sensor identification methods, it has higher environmental identification accuracy and adaptability to complex environments.

[0039] (3) The present invention utilizes a sliding time window to construct environmental features, effectively reducing the impact of instantaneous measurement noise and random disturbances on the environmental identification results and improving the stability of environmental identification.

[0040] (4) The present invention uses a multilayer perceptron environment recognition model to establish a complex nonlinear mapping relationship between multiple source sensors. Compared with traditional threshold judgment methods and traditional machine learning methods, it can more accurately distinguish between three motion environments: air, ground and water, and improve the environmental recognition performance in complex environments.

[0041] (5) The present invention introduces an environmental credibility assessment mechanism to perform reliability analysis on each environmental identification result. When the environmental characteristics are unclear or the environment is in a transitional stage, the current motion mode is maintained, which effectively reduces the error mode switching caused by misjudgment.

[0042] (6) The present invention proposes a continuous window consistency confirmation mechanism, which allows the execution mode to switch only when multiple consecutive time windows are identified as the same environment category, effectively suppressing the jump of identification results near the environment boundary, avoiding frequent switching of the control system, and improving the overall operational stability of the robot.

[0043] (7) The present invention can effectively suppress the false triggering of water level sensors caused by complex environments such as rain, fog, water vapor, water droplet adhesion and splashing waves, improve the robustness of the environmental identification system to complex environmental interference and abnormal sensor input, and improve the reliability of robot autonomous operation in complex environments.

[0044] (8) The present invention establishes an integrated autonomous control process of environmental recognition, environmental judgment, mode switching and controller switching, which enables the robot to automatically call the corresponding control strategy according to environmental changes and improve the robot's ability to operate continuously across environments.

[0045] (9) The algorithm of this invention has low computational load and high real-time performance. It can be directly deployed on embedded controllers to realize online real-time environment recognition and motion mode switching control, and has good engineering application value.

[0046] (10) This invention is not only applicable to the amphibious robot proposed in this application, but also to multimodal robots, amphibious robots and other intelligent mobile robots that have the need to switch between two or more motion modes. It has good versatility and promotional application value.

[0047] Furthermore, refer to Figure 1 Step S100 includes: S110. After the robot is powered on, the main controller initializes the inertial measurement unit, water level sensor, and real-time data buffer queue. S120. The main controller acquires data at a fixed sampling frequency, and obtains the following data X in each sampling cycle. t : X t = [a x , a y , a z , g x , g y , g z , W] Among them, a x , a y , a z For the robot's acceleration in three linear directions, g x , g y , g z The values ​​represent the robot's angular velocities in three directions, and W represents the state of the water level sensor. When W=0, it indicates that the robot is not in contact with the liquid, and when W=1, it indicates that the liquid has been detected. S130. To ensure the consistency of data from multiple sensor sources, the main controller uses a unified timestamp mechanism to synchronize all sensor data. That is, each time a sampling is completed, a sampling vector X is formed. t The data is then stored in a cyclic real-time data cache queue in chronological order for subsequent sliding window feature extraction.

[0048] In step S130, to ensure the consistency of multi-source sensor data, the main controller uses a unified timestamp mechanism to synchronize all sensor data. Each time sampling is completed, a sampling vector X is formed. t The data is then stored in a circular buffer in chronological order. The buffer length is preferably set to 1 second of historical data for subsequent sliding window feature extraction.

[0049] Furthermore, refer to Figure 1 and Figure 5 Step S200 includes: S210. Based on the preset window length and window step size, the main controller uses a sliding time window to process continuous data; Preferably, the window length T = 0.5 s and the window step size is 0.1 s. Since the sampling frequency is 100 Hz, each window contains 50 sampling points. When the robot is running, the window slides forward continuously at a fixed step size to achieve continuous updates of the environmental state.

[0050] S220: Calculates an eight-dimensional environmental feature vector F using an inertial measurement unit and a water level sensor to simultaneously reflect information such as robot posture changes, motion intensity, and whether the robot is in contact with the water environment. F = [μ ax , μ ay , μ az RMS a , μ g RMS g E a , W], Where, μ ax , μ ay , μ az RMS is the mean of the three-axis accelerations. a Let μ be the root mean square of the acceleration. g The mean angular velocity, RMS g , where E is the root mean square of the angular velocity. a W represents the acceleration energy, and W represents the state of the water level sensor.

[0051] Within each sliding window, statistical analysis is performed on the IMU data and water level sensor data to construct an environmental feature vector. Referring to Table 1, this embodiment preferably uses the following eight-dimensional features: Table 1:

[0052] Furthermore, refer to Figure 1 In step S220, the eight-dimensional environmental feature vector includes the following components: Mean triaxial acceleration μ ax μ ay and μ az Root Mean Square Acceleration (RMS) a Angular velocity mean μ g Angular velocity root mean square (RMS) g Acceleration energy E a and the water level sensor status W, where: μ ax = (1 / N) Σ a x (i), μ ay = (1 / N) Σ a y (i), μ az = (1 / N) Σ a z (i), RMS a = √[(1 / N) Σ (a x 2 + a y 2 + a z 2 )], μ g = (1 / N) Σ √(g x 2 + g y 2 + g z 2 ), RMS g = √[(1 / N) Σ (g x 2 + g y 2 + g z 2 )], E a = (1 / N) Σ (a x 2 + a y 2 + a z 2 ), Where N is the number of sampling points within a single sliding event window, a x (i), a y (i) and a z (i) represent the accelerations along the x, y, and z axes respectively during a single sampling, g x g y and g zω represents the angular velocities of the x-axis, y-axis, and z-axis in a single sampling, respectively, and the water level sensor state W is the mode of the water level sensor within the sliding time window.

[0053] Furthermore, refer to Figure 1 and Figure 6 In step S300, the Multi-Layer Perceptron (MLP) environment recognition model includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer connected in sequence. The input layer receives environmental feature vectors. The first, second, and third hidden layers employ a fully connected neural network structure and utilize a nonlinear activation function to establish a nonlinear mapping relationship between different features, thereby improving the separability between different motion states in complex environments. Specifically, the input layer has 8 nodes, the first hidden layer has 128 nodes, the second hidden layer has 64 nodes, and the third hidden layer has 32 nodes. The first, second, and third hidden layers use the ReLU activation function. ReLU(x) = max(0, x) The output layer has three nodes, and it outputs the probability value P of the robot belonging to the aerial environment via the Softmax function. air The probability value P of the ground environment ground and the probability value P of the aquatic environment water ,satisfy P air + P ground + P water = 1; The main controller selects the environment with the highest probability as the initial recognition result for the current window, denoted as . ,in, .

[0054] Specifically, in step S300, the environmental recognition model is trained offline using a large amount of sensor data collected by the robot during its flight, ground movement and water navigation. The model is also trained using supervised learning methods to establish a mapping relationship between environmental features and environmental categories, enabling the model to learn the intrinsic relationship between inertial information, water level information and motion features under different environments.

[0055] The main controller selects the environment with the highest probability as the current recognition result. For example: P air = 0.92, P ground =0.06, P water = 0.02, then the robot is currently in an aerial environment.

[0056] Compared to traditional threshold judgment methods, this invention utilizes neural networks to establish nonlinear relationships between multiple sensors. This fully leverages the complementary features between inertial measurement data and water level detection information, effectively reducing environmental misjudgments caused by anomalies in a single sensor and improving the accuracy and generalization ability of environmental recognition in complex environments. Furthermore, when the robot is subjected to external disturbances or rapid environmental changes, the multilayer perceptron can make a comprehensive judgment based on the motion state reflected by multiple sensors, rather than relying on the output of a single sensor. Therefore, it can improve the robot's autonomous recognition capability in complex environments.

[0057] Furthermore, in step S400, the maximum category probability is used as the environmental credibility Conf, i.e. Where Conf represents the confidence level of the current window environment. When the environment confidence level Conf is greater than or equal to the confidence level threshold, the confidence level is determined. At that time, that is If the current environment is reliably identified, then the robot is either still in the transition phase of the environment or is subject to strong external interference.

[0058] In some embodiments, the confidence threshold The preferred setting is: Specifically, when the robot is still in the environmental transition phase or when the robot is subjected to strong external interference, the robot maintains the current control mode, does not perform motion mode switching, and continues to collect subsequent sensor data for the next environmental recognition.

[0059] Furthermore, refer to Figure 1 In step S500, the initial recognition results of the three most recent sliding windows are as follows: The corresponding credibility levels are as follows: When the recognition results of three consecutive windows are consistent, that is: Furthermore, the environmental confidence levels of the two most recent windows are both greater than or equal to the confidence threshold. ,Right now: It is determined that the robot has stably entered the target environment.

[0060] For example, when a robot slowly descends from the air to the ground, because the robot still has a certain vertical velocity and the landing gear has already touched the ground, its inertial characteristics may simultaneously exhibit both flight and ground motion characteristics. In this case, the MLP model might output: Air: 0.53, Ground: 0.45, Water: 0.02 Although the Air category is slightly higher than the Ground category, the environmental confidence level is only 0.53, which is far below the confidence threshold. Therefore, the system will not immediately switch motion modes, but will continue to collect data from subsequent windows.

[0061] Similarly, when the robot flies in dense fog, rain, or areas with splashing waves, the water droplets in the air may cause the water level sensor to output 1, but the IMU still exhibits a stable flight state. In this case, the environment classification model can still comprehensively determine that the robot is in a flight environment, so the probability of a water environment will not suddenly increase, thus effectively avoiding the controller from switching erroneously due to false triggering of the water level sensor. Therefore, this embodiment utilizes an environmental reliability mechanism to make the robot's environment recognition more stable and improve its robustness in complex environments.

[0062] When three or more consecutive time windows identify the same environment category, and the corresponding environment confidence level meets the preset requirements, the robot is determined to have stably entered the target environment, and motion mode switching is allowed. If the identification results of multiple consecutive time windows are inconsistent, or the environment confidence level is lower than the preset threshold, the robot is considered to still be in the environment transition phase, and the system continues to maintain the current motion mode without executing controller switching.

[0063] For example: Window 1: Air; Window 2: Air; Window 3: Air, and the confidence levels of the three windows are 0.95, 0.97, and 0.96 respectively, then switching to airplane mode is allowed.

[0064] If the identification results are Air, Ground, and Ground respectively, it indicates that the robot is still in the environment transition phase, and the original movement mode should be maintained. For example, if Window 1: Water; Window 2: Water; Window 3: Air, it indicates that there is a disturbance in the current environment, and switching is not allowed. Continuous window confirmation effectively filters out momentary misjudgments caused by robot vibration, random sensor noise, water droplet adhesion, etc. Experimental results in this embodiment show that after introducing continuous window confirmation, even if the water level sensor experiences a short-term false trigger, the system can still maintain correct environmental identification, significantly improving the stability of the robot's autonomous control.

[0065] Furthermore, refer to Figure 1 In step S600, When the robot recognizes the environment as a flight environment, the system automatically enters flight mode, calls flight control parameters, and starts the flight controller. The flight controller controls the four rotors to output thrust, while shutting down the wheel drive motors and water surface propulsion, thus achieving attitude control and flight motion control. When the robot recognizes the ground environment, the system automatically switches to ground motion mode, turns off flight propulsion output, calls differential control parameters, and starts the wheel motion controller and left and right wheel differential controller to realize the robot's ground movement; When the robot recognizes a water surface environment, the system automatically switches to water surface motion mode, calls up water surface navigation control parameters, shuts down the flight controller, and starts the water surface propulsion controller to enable the robot to navigate on the water surface.

[0066] During the motion mode switching process, the control system simultaneously completes the switching of actuator output, control parameters, and corresponding control algorithms, enabling the robot to automatically call the corresponding control strategy according to different motion environments and complete autonomous operation across environments without human intervention.

[0067] Furthermore, refer to Figure 1 In step S600, the mode switching control module is further equipped with a motion mode switching protection mechanism, which includes: a minimum hold time limit, an actuator switching sequence limit to ensure that the robot does not conflict due to simultaneous operation of the controller, and an abnormal recovery mechanism. The minimum holding time limit is the minimum preset holding time after the robot completes a mode switch. During the minimum holding time, switching again is prohibited to avoid the robot repeatedly entering different control modes. The actuator switching sequence restrictions include restrictions on switching from flight mode to ground mode, ground mode to flight mode, and flight mode to water surface mode. The restrictions for switching from flight mode to ground mode include first reducing the speed of the flight motors, confirming that the aircraft has touched the ground, then turning off flight control, and finally activating wheel control. The restrictions for switching from ground mode to flight mode include first activating the flight controller to stabilize the attitude, then activating the flight motors, and finally disabling the wheel drive. Switching from flight mode to water surface mode has the following restrictions: first, reduce altitude to confirm contact with the water surface, then turn off the flight controllers, and finally activate the thrusters.

[0068] The anomaly recovery mechanism ensures that if the credibility requirement is not met for any consecutive pre-screening windows during the environment identification process, the system maintains the current operating mode and re-enters the environment identification process. This avoids frequent controller switching due to drastic environmental changes.

[0069] In some embodiments, step S600 further includes adding conditions such as height constraints, posture constraints, speed constraints, and task state constraints to the motion mode switching according to the actual task requirements of the robot, so as to achieve safer and more reliable cross-environment autonomous control.

[0070] Furthermore, refer to Figure 2 The present invention also proposes a triphibious robot for performing the autonomous environment recognition and cross-domain movement mode switching method of the triphibious robot. The triphibious robot has the ability to fly in the air, move on the ground, and navigate on the water. The triphibious robot includes: The robot body includes an actuator and control module and a power management module. The actuator and control module includes at least a flight drive mechanism, a ground movement mechanism, and a water propulsion mechanism. A multi-source sensor module is also included on the robot body. This multi-source sensor module includes at least an inertial measurement unit and a water level sensor. The inertial measurement unit (IMU) is a six-axis IMU used to acquire the robot's motion state information in real time. The IMU is installed near the center of the robot body and is used to measure the robot's three-axis linear acceleration and three-axis angular velocity information in real time. The IMU outputs the robot's linear acceleration and angular velocity in three directions, namely the x-axis acceleration, y-axis acceleration, z-axis acceleration, and angular velocity around the x-axis, y-axis, and z-axis, respectively. The water level sensor is used to detect whether the robot is in contact with water. The water level sensor is installed on the bottom of the robot and outputs the detection result of whether the robot is currently in contact with water. The main controller is used to complete environmental recognition calculations and motion mode decisions. The multi-source sensor module is electrically connected to the main controller. The main controller completes the time synchronization and caching of data from each sensor, providing a unified data input for subsequent environmental recognition. The main controller uses an embedded processor as the core computing platform to complete functions such as data acquisition, environmental recognition, motion mode determination, and controller switching. The main controller is equipped with an environment recognition module and a motion mode switching control module. The environment recognition module uses a trained environment classification model to determine the robot's current environment in real time based on inertial measurement data and water level sensor data. The motion mode switching module manages the flight controller, ground controller, and water surface controller in a unified manner based on the output results of the environment recognition module, and automatically completes the controller switching.

[0071] Reference Figure 3To realize the method for autonomous environment recognition and cross-domain motion mode switching of the amphibious robot proposed in this invention, this embodiment constructs a complete robot hardware system. This system uses the robot body as a carrier and integrates multiple functional modules such as flight drive, ground drive, environmental perception, attitude measurement, navigation and positioning, and embedded computing. These functional modules work together to complete the robot's autonomous environment recognition, motion control, and mode switching tasks.

[0072] In this embodiment, the robot body mainly includes hardware units such as a lithium battery, a four-in-one electronic speed controller, a brushless motor, a propeller, a DC geared motor, a water level sensor, an ESP32 controller, an onboard computer, a flight controller, a global positioning system (GPS) module, and servo motors.

[0073] The lithium battery serves as the power supply unit for the entire robot system, providing a stable power source for the flight drive system, ground drive system, control system, and sensor system. The four-in-one electronic speed controller connects to four brushless motors and adjusts the speed of each brushless motor according to the control commands output by the flight controller, thereby achieving robot attitude control and flight power output. The brushless motors drive the propellers to generate lift, enabling the robot to perform flight maneuvers such as takeoff, hovering, flying, and landing.

[0074] DC geared motors are installed on the robot's wheels to drive the ground movement mechanism, enabling the robot to move forward, backward, and turn on land. When the robot enters ground movement mode, the flight propulsion system stops outputting power, and the DC geared motors take over the ground movement task, improving the robot's ground movement efficiency and reducing the energy consumption of the flight system. Simultaneously, DC geared motors on the wheels are also used to drive movement on water. In wheel mode, the DC geared motors change direction to achieve passive transformation, transforming the robot into a propeller mode for propulsion on water.

[0075] A water level sensor is installed at the front of the robot to detect in real time whether the robot is in contact with liquid. When the robot enters a watery environment, the water level sensor can quickly output the liquid contact status, providing important information for environmental identification. Furthermore, this invention does not directly rely on the water level sensor output to switch modes, but rather fuses it with inertial measurement information, thereby improving the reliability of environmental identification in complex environments.

[0076] The ESP32 controller is mainly responsible for the robot's underlying drive control and some sensor data acquisition, and completes data interaction with peripheral devices such as water level sensors, DC motors, and servo motors; the flight controller is used to complete flight attitude calculation, inertial measurement unit data processing, and flight control algorithm operation, and outputs robot attitude information, linear acceleration information, and angular velocity information in real time; the global positioning system module is used to provide robot position and velocity information, providing auxiliary positioning information for robot navigation and flight control; the servo motors are used to drive the robot's corresponding actuators to complete mechanical actions.

[0077] The onboard computer serves as the robot's host computing platform, responsible for running the autonomous environment recognition algorithm, motion mode switching algorithm, and robot control program proposed in this invention. It also establishes data communication with the flight controller and ESP32 controller to achieve functions such as multi-source sensor data fusion, environmental recognition, and control decision-making.

[0078] The aforementioned hardware modules are connected via a communication bus to form a complete robot control system, enabling functions such as robot environmental perception, data processing, motion control, and mode switching, thus providing hardware support for the autonomous environment recognition method proposed in this invention.

[0079] Reference Figure 4 This embodiment further establishes a robot control software system. The entire software system adopts a layered control architecture, consisting of an operating system layer, a robot software platform layer, a motion mode management layer, and a human-machine interaction layer, realizing functions such as robot environment recognition, mode management, motion control, and ground monitoring.

[0080] The robot's onboard computer runs the Ubuntu operating system, and the Robot Operating System (ROS) software platform is deployed on the Ubuntu system. The Ubuntu operating system provides a stable software operating environment for the robot, and ROS is responsible for data communication, node management, message transmission, and control task scheduling between the robot's various functional modules, enabling the robot's various control modules to achieve modular development and real-time collaborative work.

[0081] The autonomous environment recognition program and motion mode management program proposed in this invention are run on the ROS platform. After the robot collects data from the inertial measurement unit, water level sensor and other sensors in real time, the autonomous environment recognition program first completes data preprocessing, sliding window construction, environmental feature extraction and MLP neural network classification, and outputs the current environment category of the robot; then the motion mode management module automatically switches the corresponding motion mode according to the environment recognition results.

[0082] In this embodiment, the motion modes mainly include three working states: aerial mode, ground mode, and water surface mode. In aerial mode, the robot invokes the flight control program to achieve attitude control and flight maneuvers; in ground mode, the system automatically switches to the ground motion control program to achieve wheeled movement of the robot; in water surface mode, the control system invokes the water surface propulsion control program to achieve autonomous navigation of the robot in the water environment. The three motion modes are scheduled by a unified mode management module and automatically switch according to the environmental recognition results proposed in this invention, without manual intervention.

[0083] To enable remote monitoring and task management of the robot, this embodiment further establishes a wireless communication network to achieve real-time data interaction between the robot and the ground station. The robot transmits its flight status, motion status, environmental recognition results, and system operating status to the ground station in real time via the wireless network. The ground station can display the robot's current motion mode, position, attitude, and operating status in real time, and send control commands or task planning information to the robot as needed.

[0084] The wireless communication link is mainly used for information exchange between the robot and the ground station, including robot posture information, environmental recognition status, flight status, movement status, and control commands. The ground station can monitor the robot's operating status in real time and also serve as an interface for human intervention, sending control commands to the robot in special circumstances to achieve flexible switching between autonomous and manual control.

[0085] Through the collaborative work of the aforementioned software and hardware systems, this invention establishes a complete autonomous control process from environmental information acquisition, environmental identification, motion mode management, and controller switching to robot actuator control. This enables the robot to achieve autonomous identification, autonomous decision-making, and autonomous motion mode switching in three environments: air, ground, and water. It provides complete software and hardware support for the autonomous operation of robots in complex environments.

[0086] Reference Figure 7 and Figure 8 To verify the effectiveness of the autonomous environment recognition and motion mode switching method proposed in this invention, environmental recognition experiments were conducted under two conditions: no environmental interference and environmental interference.

[0087] Reference Figure 7 During the experiment, the robot collected test data in three typical environments: air flight, ground movement, and water navigation, and input the data into the environmental recognition model proposed in this invention for classification. Figure 8As can be seen, in the aerial environment test samples, 293 samples were correctly identified as aerial environments, and only 5 samples were misidentified as ground environments; in the ground environment test samples, 298 samples were correctly identified, and only 7 samples were misidentified as aerial environments; all 195 samples in the water surface environment test samples were correctly identified, with no misclassifications. The overall recognition accuracy reached over 97.5%, indicating that the multi-source sensor fusion environment recognition method proposed in this invention can effectively distinguish between aerial, ground, and water surface motion environments, and has high environment classification accuracy.

[0088] To further verify the robustness of this invention in complex environments, environmental disturbances were artificially introduced during flight to simulate conditions such as rain, fog, high humidity, or splashing waves. Water mist, droplets, and vapor in the air adhered to the surface of the water level sensor, causing the sensor's output state to change from 0 to 1, thus simulating a false triggering of the water level sensor. The same data acquisition and recognition process was used to obtain... Figure 8 The results are shown.

[0089] Reference Figure 8 After adding environmental interference, 289 samples of the aerial environment were correctly identified, with only 7 samples misidentified as ground environments and 2 samples misidentified as water environments; 297 samples of the ground environment were correctly identified, with only 8 samples misidentified as aerial environments; and all 195 samples of the water environment remained correctly identified, with no misclassification. Compared to the absence of interference, although the accuracy of aerial environment identification decreased slightly, the overall identification performance remained stable, with only a very small increase in misclassified samples. This indicates that the environmental identification model proposed in this invention does not directly classify the robot as being in a water environment due to a single false triggering of the water level sensor, but rather can comprehensively combine inertial measurement information and water level sensor information for joint judgment, thereby effectively suppressing misidentification caused by factors such as rain, fog, water vapor, and water droplet adhesion.

[0090] Further analysis Figure 7 and Figure 8 It can be observed that, under both experimental conditions, misclassification mainly occurred between the aerial and ground environments, while the water environment consistently maintained a high recognition accuracy. This is because the robot's acceleration and attitude changes during its descent from the air to the ground are somewhat similar to those during the ground launch phase, making misclassification between air and ground scenarios more likely in a small number of samples. The water environment, however, not only possesses significantly different inertial characteristics but also incorporates water level sensor information, thus exhibiting better separability. Even with the addition of artificial interference from the water level sensor, the multilayer perceptron classification model proposed in this invention can still identify that the robot is still in a flight environment based on the IMU's motion state, thereby avoiding erroneous mode switching caused by relying solely on a single water level sensor for environmental judgment.

[0091] Experimental results show that the autonomous environment recognition method proposed in this invention can fully integrate multi-source information such as inertial measurement unit and water level sensor, and has high environmental recognition accuracy and strong anti-interference ability. Combined with environmental reliability assessment and continuous window consistency confirmation mechanism, it can effectively avoid environmental misjudgment caused by sensor false triggering, environmental noise and complex working conditions, providing a reliable basis for the robot to switch autonomous motion modes across environments, thereby improving the stability and safety of the amphibious robot's autonomous operation in complex environments.

[0092] The practical application process of this embodiment is illustrated by taking the robot performing the task of "air flight - autonomous landing - ground movement - entering the water surface - water surface navigation" as an example.

[0093] After the robot takes off, the IMU detects that the robot is in a stable flight state. The MLP model continuously outputs an aerial environment probability greater than 0.98. After confirmation through three consecutive sliding windows, the system maintains the flight mode.

[0094] As the robot autonomously lands on the ground, its linear acceleration and angular velocity, measured by the IMU, change with decreasing altitude, causing the environment classification model to output a higher probability of the ground environment. Within several consecutive sliding windows after the robot contacts the ground, the ground environment probability remains consistently higher than a preset confidence threshold. The continuous window consistency confirmation mechanism determines that the robot has stably entered the ground environment, and the main controller shuts down the flight controller and activates the ground differential controller, enabling the robot to autonomously switch to ground movement mode.

[0095] Subsequently, the robot moves to the shore and enters the water following a preset path. Upon contact with the water, the water level sensor output changes from 0 to 1, and the motion characteristics measured by the IMU gradually show a floating state. The environmental classification model comprehensively analyzes information from multiple sensor sources, continuously outputting the water surface environment as the highest probability category. After confirmation via a continuous window, the system shuts down the ground drive mechanism and activates the water surface propulsion controller, enabling the robot to enter water surface navigation mode.

[0096] Throughout the entire cross-environment operation, no manual intervention from operators is required. The robot can autonomously complete environmental recognition, motion mode determination, and controller switching. When encountering interference such as rain, fog, water vapor, or splashing waves during flight, even if the water level sensor is briefly mistriggered, the MLP model can still output the correct environmental category based on the IMU motion status. The continuous window confirmation mechanism further suppresses mode switching caused by momentary misjudgment, thereby ensuring the stability and reliability of the robot control system.

[0097] The present invention also proposes a computer device, including a memory and a processor, wherein the processor executes a computer program stored in the memory to implement the aforementioned method for autonomous environment recognition and cross-domain motion mode switching of the amphibious robot.

[0098] The present invention also proposes a computer-readable storage medium storing program instructions thereon, which, when executed by a processor, implement the method for autonomous environment recognition and cross-domain motion mode switching of the amphibious robot.

[0099] It should be understood that the method steps in all embodiments of the present invention can be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium. The methods can use standard programming techniques. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if necessary, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit (ASIC).

[0100] Furthermore, the procedures described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by the context. The procedures described herein (or variations and / or combinations thereof) may be executed under the control of one or more computer systems configured with executable instructions, and may be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program comprises a plurality of instructions executable by one or more processors.

[0101] Furthermore, the method can be implemented in any suitable type of computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices, etc. Aspects of the invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it is readable by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein. Furthermore, the machine-readable code, or portions thereof, can be transmitted via wired or wireless networks. The invention described herein includes these and other different types of non-transitory computer-readable storage media when such media comprises instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. When programmed according to the methods and techniques described in the invention, the invention may also include the computer itself.

[0102] A computer program can be applied to input data to perform the functions described herein, thereby transforming the input data to generate output data stored in non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including specific visual depictions of physical and tangible objects generated on the display.

[0103] The above description is merely a preferred embodiment of the present invention. The present invention is not limited to the above-described embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention, as long as they achieve the technical effects of the present invention by the same means, should be included within the scope of protection of the present invention. Within the scope of protection of the present invention, the technical solutions and / or implementation methods can have various modifications and variations.

Claims

1. A method for autonomous environment recognition and cross-domain motion mode switching of an amphibious robot, wherein the method is applied to an amphibious robot, the amphibious robot comprising at least a robot body equipped with a flight drive mechanism, a ground movement mechanism, and a water surface propulsion mechanism, a multi-source sensor module and a main controller mounted on the robot body, the multi-source sensor including at least an inertial measurement unit and a water level sensor, characterized in that, The method for autonomous environment recognition and cross-domain motion mode switching of the amphibious robot includes the following steps: S100: The main controller synchronously collects the output data of the multi-source sensors according to the preset sampling frequency, and uses a unified timestamp method to synchronize the output data of the multi-source sensors and establish a real-time data cache queue. S200: The main controller constructs a sliding time window based on the continuously acquired multi-source sensor data according to a preset time length, and extracts features from the data within each sliding time window. Within each sliding time window, it fuses the output data of the inertial measurement unit and the state of the water level sensor to construct an environmental feature vector that reflects the robot's current motion environment. Step S200 includes: S210. Based on the preset window length and window step size, the main controller uses a sliding time window to process continuous data; S220: Calculates an eight-dimensional environmental feature vector F using an inertial measurement unit and a water level sensor to simultaneously reflect information such as robot posture changes, motion intensity, and whether the robot is in contact with the water environment. F = [μ ax , m ay , m az , RMS a , m g , RMS g , E a , W], Where, μ ax , μ ay , μ az RMS is the mean of the three-axis accelerations. a Let μ be the root mean square of the acceleration. g The mean angular velocity, RMS g , where E is the root mean square of the angular velocity. a W represents the acceleration energy, and W represents the state of the water level sensor. In step S220, the eight-dimensional environmental feature vector includes the following components: Mean triaxial acceleration μ ax μ ay and μ az Root Mean Square Acceleration (RMS) a Angular velocity mean μ g Angular velocity root mean square (RMS) g Acceleration energy E a and the water level sensor status W, where: m ax = (1 / N) Σ a x (i), m ay = (1 / N) Σ a y (i), m az = (1 / N) Σ a z (i), RMS a = √[(1 / N) Σ (a x 2 + a y 2 + a z 2 )], m g = (1 / N) Σ √(g x 2 + g y 2 + g z 2 ), RMS g = √[(1 / N) Σ (g x 2 + g y 2 + g z 2 )], AND a = (1 / N) Σ (a x 2 + a y 2 + a z 2 ), Where N is the number of sampling points within a single sliding event window, a x (i), a y (i) and a z (i) represent the accelerations along the x, y, and z axes respectively during a single sampling, g x g y and g z These represent the angular velocities along the x, y, and z axes respectively during a single sampling. The water level sensor state W is the mode of the water level sensor within the sliding time window. x , a y , a z Accelerate the robot along three lines; S300: The main controller inputs the environmental feature vector into the pre-trained multi-layer perceptron (MLP) environmental recognition model to classify and identify the robot's current environment. In step S300, the MLP environmental recognition model includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer connected in sequence. The input layer receives environmental feature vectors. The first, second, and third hidden layers employ a fully connected neural network structure and utilize a nonlinear activation function to establish a nonlinear mapping relationship between different features, thereby improving the separability between different motion states in complex environments. Specifically, the input layer has 8 nodes, the first hidden layer has 128 nodes, the second hidden layer has 64 nodes, and the third hidden layer has 32 nodes. The first, second, and third hidden layers use the ReLU activation function. The output layer has three nodes, and it outputs the probability value P of the robot belonging to the aerial environment via the Softmax function. air The probability value P of the ground environment ground and the probability value P of the aquatic environment water ,satisfy P air + P ground + P water = 1; The main controller selects the environment with the highest probability as the initial recognition result for the current window, denoted as . ,in, ； S400 After the multilayer perceptron environment recognition model completes the classification, it outputs the predicted probability of the robot belonging to three categories: aerial environment, ground environment, and water environment. The main controller determines the most likely environment where the robot is currently located based on the predicted probability corresponding to the three categories, and further calculates the credibility of the environment recognition. S500: The main controller continuously saves the environment recognition results corresponding to multiple sliding time windows, and determines whether the environment recognition results are consistent within multiple consecutive time windows. Based on at least one latest environment recognition reliability assessment result, it determines whether the robot has stably entered the target environment. In step S500, the initial recognition results of the three most recent sliding windows are as follows: The corresponding credibility levels are as follows: When the recognition results of three consecutive windows are consistent, that is: Furthermore, the environmental confidence levels of the two most recent windows are both greater than or equal to the confidence threshold. ,Right now: The robot has been determined to have stably entered the target environment. S600: Once it is determined that the robot has stably entered the target environment, the mode switching control module automatically completes the corresponding motion mode switching based on the environment recognition result. S700, cyclically execute steps S100 to S600, continuously collect robot operating status information, and cyclically execute environment recognition, environment judgment and motion mode switching control according to the above steps.

2. The method for autonomous environment recognition and cross-domain motion mode switching of a triphibious robot according to claim 1, characterized in that, Step S100 includes: S110. After the robot is powered on, the main controller initializes the inertial measurement unit, water level sensor, and real-time data buffer queue. S120. The main controller acquires data at a fixed sampling frequency, and obtains the following data X in each sampling cycle. t : X t = [a x , a y , a z , g x , g y , g z , W] Among them, a x , a y , a z For the robot's acceleration in three linear directions, g x , g y , g z The values ​​represent the robot's angular velocities in three directions, and W represents the state of the water level sensor. When W=0, it indicates that the robot is not in contact with the liquid, and when W=1, it indicates that the liquid has been detected. S130. To ensure the consistency of data from multiple sensor sources, the main controller uses a unified timestamp mechanism to synchronize all sensor data. That is, each time a sampling is completed, a sampling vector X is formed. t The data is then stored in a cyclic real-time data cache queue in chronological order for subsequent sliding window feature extraction.

3. The method for autonomous environment recognition and cross-domain motion mode switching of a triphibious robot according to claim 1, characterized in that, In step S400, the maximum category probability is used as the environmental credibility Conf, i.e. Wherein, Conf represents the confidence level of the current window environment recognition; When the environmental credibility Conf is greater than or equal to the credibility threshold At that time, that is If the current environment is reliably identified, then the robot is either still in the transition phase of the environment or is subject to strong external interference.

4. The method for autonomous environment recognition and cross-domain motion mode switching of a triphibious robot according to claim 1, characterized in that, In step S600, When the robot recognizes the environment as a flight environment, the system automatically enters flight mode, calls flight control parameters, and starts the flight controller. The flight controller controls the four rotors to output thrust, while shutting down the wheel drive motors and water surface propulsion, thus achieving attitude control and flight motion control. When the robot recognizes the ground environment, the system automatically switches to ground motion mode, turns off flight propulsion output, calls differential control parameters, and starts the wheel motion controller and left and right wheel differential controller to realize the robot's ground movement; When the robot recognizes a water surface environment, the system automatically switches to water surface movement mode, calls water surface navigation control parameters, shuts down the flight controller, and starts the water surface propulsion controller to enable the robot to navigate on the water surface. During the motion mode switching process, the control system simultaneously completes the switching of actuator output, control parameters, and corresponding control algorithms, enabling the robot to automatically call the corresponding control strategy according to different motion environments and complete autonomous operation across environments without human intervention.

5. The method for autonomous environment recognition and cross-domain motion mode switching of a triphibious robot according to claim 4, characterized in that, In step S600, the mode switching control module is further equipped with a motion mode switching protection mechanism, which includes: a minimum hold time limit, an actuator switching sequence limit to ensure that the robot does not conflict due to simultaneous operation of the controller, and an abnormal recovery mechanism. The minimum holding time limit is the minimum preset holding time after the robot completes a mode switch. During the minimum holding time, switching again is prohibited to avoid the robot repeatedly entering different control modes. The actuator switching sequence restrictions include restrictions on switching from flight mode to ground mode, ground mode to flight mode, and flight mode to water surface mode. The restrictions for switching from flight mode to ground mode include first reducing the speed of the flight motors, confirming that the aircraft has touched the ground, then turning off flight control, and finally activating wheel control. The restrictions for switching from ground mode to flight mode include first activating the flight controller to stabilize the attitude, then activating the flight motors, and finally disabling the wheel drive. Switching from flight mode to water surface mode has the following restrictions: first, reduce altitude to confirm contact with the water surface, then turn off the flight controller, and finally activate the thrusters. The anomaly recovery mechanism is that when the number of consecutive pre-screening windows fails to meet the credibility requirements during the environment identification process, the system maintains the current mode and re-enters the environment identification process to avoid frequent controller switching due to drastic environmental changes.

6. A triphibious robot, characterized in that, A method for performing autonomous environment recognition and cross-domain motion mode switching of an amphibious robot as described in any one of claims 1 to 5, wherein the amphibious robot has the capabilities of aerial flight, ground movement, and water navigation, and the amphibious robot comprises: The robot body is equipped with an actuator and control module and a power management module. The actuator and control module includes at least a flight drive mechanism, a ground movement mechanism and a water propulsion mechanism. A multi-source sensor module is mounted on the robot body. The multi-source sensor module includes at least an inertial measurement unit and a water level sensor. The inertial measurement unit (IMU) is installed near the center of the robot body and is used to measure the robot's three-axis acceleration and three-axis angular velocity information in real time. The IMU outputs the linear acceleration and angular velocity of the robot in three directions, namely the x-axis acceleration, y-axis acceleration, z-axis acceleration, and angular velocity around the x-axis, y-axis, and z-axis, respectively. The water level sensor is used to detect whether the robot is in contact with water and is installed on the bottom of the robot. The main controller is used to perform environmental recognition calculations and motion mode decisions. The multi-source sensor module is electrically connected to the main controller. The main controller is equipped with an environmental recognition module and a motion mode switching control module. The environmental recognition module uses a trained environmental classification model to determine the robot's current environment in real time based on inertial measurement data and water level sensor data. The motion mode switching module manages the flight controller, ground controller, and water surface controller in a unified manner based on the output results of the environmental recognition module, and automatically completes the controller switching.

Images