Decoupling control method for motion coupling between track and propeller

Abstract

The invention discloses a decoupling control method for motion coupling between a track and a propeller, relates to the technical field of robot control, and realizes accurate and intelligent regulation and control of motion of a sea-land amphibious robot by constructing a complete closed-loop strategy from sensing, recognition, prediction to control. And the motion stability of the robot in a complex mixed environment is obviously improved. According to the method, the undesirable coupling effect between the crawler belt and the propeller can be anticipated and quantified in a prospective mode, suppression is conducted actively through self-adaptive control, impact, vibration and attitude instability caused by power conflicts in land and water transition or rugged terrain are effectively avoided, and stability and safety of a robot platform and loads of the robot platform are guaranteed. The energy utilization efficiency of the robot is greatly improved. And it is ensured that the two power systems work in the optimal cooperative state at any moment, and invalid energy internal consumption and redundant output are avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of robot control technology, in particular to a decoupling control method for motion coupling between a track and a propeller. BACKGROUND

[0002] An amphibious tracked robot is a special mobile platform capable of performing tasks in both land and water environments. This type of robot usually integrates two independent power systems, namely a tracked system for land travel and a propeller system for water navigation. In order to achieve full-terrain maneuverability, how to effectively manage and control these two power systems with completely different physical characteristics and working environments is a core technical problem that determines the overall performance.

[0003] In the prior art, the control of such amphibious robots usually adopts a simple mode switching or power superposition strategy. For example, a common control method is to use a water depth sensor or similar environmental sensor to determine when the robot enters the water and the depth exceeds a certain set threshold, the system switches control from the tracked system to the propeller system, and vice versa. Another way is to allow the tracked system and the propeller system to work simultaneously in the transition area or in a specific mode, and the power output is distributed according to a pre-set fixed ratio or simply superimposed.

[0004] However, the above prior art solutions have obvious defects in actual application. The threshold-based hard switching method often causes a significant power interruption and mechanical impact when the robot enters or exits the water, seriously affecting the smoothness of the motion and the stability of the platform. Simple power superposition or fixed ratio distribution ignores the complex fluid-solid coupling and dynamic interference effects of the tracked and propeller in mixed media, and in water-land transition zones such as swamps and shallow waters, the two systems often produce non-coordinated motion, and even counteract each other, not only leading to a decline in motion control accuracy, but also causing serious energy waste and mechanical wear. SUMMARY

[0005] The purpose of the present application is to provide a decoupling control method for motion coupling between a track and a propeller, which solves the problems in the background art.

[0006] To solve the above technical problems, the present application adopts the following technical solutions: The present application provides a decoupling control method for motion coupling between a track and a propeller, comprising: step one, acquiring multi-source sensor data and performing fusion processing on the multi-source sensor data to generate environment state data and vehicle body state data.

[0007] Step two, based on the environment state data and the vehicle body state data, identifying the current terrain type and motion mode requirement, generating a terrain identification result and a mode requirement signal.

[0008] Step three, according to the terrain recognition result and the mode demand signal, a track-propeller-vehicle body coupling dynamics model is constructed, and a rolling horizon prediction method is used to predict the coupling influence degree in the future time period, and a prediction result is generated.

[0009] Step four, based on the prediction result, an adaptive decoupling controller is used to dynamically adjust the power distribution between the track motor and the propeller, and a power distribution parameter is generated.

[0010] Step five, according to the power distribution parameter, track motor control signal and propeller control signal are outputted to realize the decoupling of motion coupling.

[0011] The beneficial effects of the present application are: the present application realizes accurate and intelligent regulation and control of the motion of amphibious robots through the construction of a complete closed-loop strategy from perception, recognition, prediction to control, which significantly improves the motion stability of robots in complex mixed environments. This method can foresee and quantify the adverse coupling effect between the track and the propeller in advance, and actively suppress it through adaptive control, effectively avoiding the impact, vibration and attitude instability caused by power conflict in water-land transition or rugged terrain, and ensuring the stability and safety of the robot platform and its load.

[0012] The present application also greatly improves the energy utilization efficiency of the robot. Unlike traditional threshold-based hard switching or simple power superposition strategy, this method accurately evaluates the power demand through a coupling dynamics model and intelligently allocates the power output of the track and the propeller, ensuring that the two power systems work in an optimal collaborative state at any time, avoiding invalid energy consumption and redundant output, thereby significantly reducing the overall energy consumption on the premise of completing the same task, effectively prolonging the operation endurance time of the robot.

[0013] In addition, the present application enhances the autonomous adaptation ability of the robot to unknown and dynamically changing environment. Through real-time optimization and learning mechanism, the control system can continuously adjust and improve its control parameters and model according to the accumulated operation data, so that it can not only cope with the preset typical working conditions, but also quickly adapt to new and unstructured environments. This self-learning and evolution ability makes the robot have higher robustness and task generalization ability, reduces the dependence on artificial intervention and pre-debugging, and improves its autonomous level and operation efficiency in practical application. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only show some of the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0015] Figure 1 The method of the present application is implemented in the flowchart. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0017] Referring to Figure 1 As shown in the drawings, the present application provides a decoupling control method for motion coupling between a track and a propeller, comprising: step one, acquiring multi-source sensor data and performing fusion processing on the multi-source sensor data to generate environment state data and vehicle body state data.

[0018] In specific embodiments of the present application, acquiring multi-source sensor data and generating environment state data and vehicle body state data comprises: collecting data in real time through multiple sensors deployed on the robot chassis to obtain water depth sensor data, terrain recognition camera data, inertial measurement unit data and motor current sensor data, forming a set of raw sensor data.

[0019] The water depth sensor is the distance from the bottom of the vehicle body to the water bottom or ground. The terrain recognition camera data is a continuous image frame reflecting the terrain and topography in front. The inertial measurement unit data is the original reading of three-axis acceleration and three-axis angular velocity. The motor current sensor data reflects the working load of the track drive motor and the propeller motor.

[0020] The set of raw sensor data is filtered and corrected for preprocessing to generate preprocessed sensor data.

[0021] Specifically, the inertial measurement unit data is processed using Kalman filtering algorithm to correct its inherent zero drift and random noise. The terrain recognition camera data is subjected to image de-warping and white balance correction to improve image quality. The water depth sensor data and motor current sensor data are subjected to sliding average filtering to smooth transient fluctuations.

[0022] inputting the pre-processed sensor data into a data fusion algorithm, outputting the environment state data and the vehicle body state data.

[0023] In particular, the data fusion algorithm is a multi-rate extended Kalman filter to generate accurate environment state data and vehicle body state data. In the algorithm, the vehicle body state is defined as a state vector containing position, velocity, attitude angle, etc., while the environment state contains terrain type, water depth, etc. Through the time update step, the state is predicted, and when receiving the pre-processed sensor data of different sensors, the state vector is corrected through the measurement update step. For example, the attitude angle estimation of the vehicle body can be represented by the following formula: wherein, is the fused attitude angle estimation, which is part of the vehicle body state data, is the attitude angle obtained by integrating the inertial measurement unit data, is the attitude angle calculated by analyzing the visual odometry information in the terrain recognition camera data, is the attitude change amount indirectly calculated by analyzing the motor current sensor data to judge the load difference of the two sides of the track, are the dynamic weight coefficients of each sensor data, and their sum is 1. The weight coefficients are dynamically adjusted according to the confidence of each sensor in the current environment. For example, when the underwater vision is blurred, will be reduced, and and will be increased accordingly. In this way, the data fusion algorithm can integrate various types of information to output high-precision and high-reliability environment state data and vehicle body state data.

[0024] This method solves the problem of insufficient reliability of a single sensor in a complex amphibious environment through collaborative processing and deep fusion of multi-source heterogeneous information such as water depth, vision, inertia, and motor load. It generates a more comprehensive, accurate, and robust robot state and environment perception result than any independent information source. This high-confidence environment state data and vehicle body state data provides a solid data foundation for subsequent coupled dynamics modeling, prediction, and decoupling control, ensuring the accuracy and effectiveness of the entire control system decision-making, thereby significantly improving the adaptability and operational stability of the robot in unknown and dynamically changing environments.

[0025] Step two, based on the environment state data and the vehicle body state data, identifying the current terrain type and motion mode requirement, generating terrain recognition results and mode requirement signals.

[0026] ​​Specifically, the process of identifying the current terrain type and motion mode requirement in the present application is a key link for converting the environment state data and the vehicle body state data generated in the previous steps into specific control instructions. The process first extracts environmental features from the environment state data to generate an environmental feature vector.

[0027] In specific embodiments of the present application, the identification of the current terrain type and motion mode requirement includes extracting water depth features and terrain texture features from the environment state data to generate an environmental feature vector.

[0028] Specifically, the real-time water depth value and its rate of change contained in the environment state data are taken as water depth features. At the same time, the image data from the terrain recognition camera in the environment state data is input into a pre-trained convolutional neural network, and the feature map output by the deep network is extracted as terrain texture features. The water depth features and the terrain texture features are spliced to form an environmental feature vector containing key information of the current external environment.

[0029] The attitude features and motion speed features are extracted from the vehicle body state data to generate a vehicle body feature vector.

[0030] Specifically, the pitch angle, roll angle and corresponding angular velocity of the robot are extracted from the vehicle body state data as attitude features. At the same time, the forward speed, lateral slip rate and speed difference of the two sides of the track are extracted as motion speed features. These attitude features and motion speed features reflecting the motion state of the robot are combined to form a vehicle body feature vector representing the current dynamic response of the robot.

[0031] The environmental feature vector and the vehicle body feature vector are input into a classification model to output the terrain recognition result and the mode requirement signal.

[0032] It should be noted that the classification model, such as a support vector machine or a random forest classifier, has learned a complex mapping relationship between the comprehensive features and the terrain and mode based on a large amount of offline training data, and its output result is two discrete signals, i.e. the terrain recognition result and the mode requirement signal. The terrain recognition result is a judgment of the current environment, such as hard land, marsh, shallow water area or deep water area. The mode requirement signal is the optimal working mode determined according to the current state, such as land driving mode, water-land transition mode or water area navigation mode. The classification process can be represented by the following function: wherein, represents the output terrain recognition result and mode requirement signal. represents the trained classification model function, represents the environmental feature vector, representing the vehicle body feature vector, representing a splicing operation on two feature vectors.

[0033] The method realizes deep understanding and accurate judgment of a complex amphibious environment by fusing external environment perception information and robot internal state response information. The feature extraction and classification strategy of internal and external combination enables the system to not only identify static terrain features, but also perceive dynamic performance when the robot interacts with the environment, thereby obtaining situation awareness capability far beyond a single information source, improving original sensor data to mode signals with clear physical meaning of terrain categories and high-level control intentions, and providing clear and accurate decision inputs for subsequent dynamic modeling and adaptive control. It is the core foundation for realizing robot intelligence, smooth mode switching and efficient operation.

[0034] Step three, according to the terrain recognition result and the mode demand signal, a tracked-propeller-vehicle body coupled dynamics model is constructed, and a rolling horizon prediction method is used to predict the coupled influence degree in a future time period to generate a prediction result.

[0035] Specifically, the process of establishing the tracked-propeller-vehicle body coupled dynamics model in the application is the core step of realizing predictive control. Firstly, according to the terrain recognition result and the mode demand signal output from the previous link, a set of model parameters most suitable for the current working condition is selected from a preset model parameter library. This selection process ensures that the dynamics model can reflect the physical interaction characteristics of the robot and the current environment in real time.

[0036] In specific embodiments of the application, the construction of the tracked-propeller-vehicle body coupled dynamics model includes: according to the terrain recognition result and the mode demand signal, a set of model parameters is matched and selected from a parameter library.

[0037] The parameter library stores physical parameters for different terrains such as hard land, marsh, water area and motion modes such as land travel, water-land transition, such as ground adhesion coefficient, soil cohesion, fluid density, hydrodynamic drag coefficient, etc.

[0038] Based on the model parameters, a dynamics equation containing fluid resistance term, tracked ground pressure term and propeller thrust nonlinear term is established. The dynamics equation comprehensively considers all main external forces and torques acting on the robot. The core is the Newton-Euler equation. The dynamics equation explicitly includes the key mechanical terms, including the fluid resistance term, the tracked ground pressure term and the propeller thrust nonlinear term. The simplified dynamics equation along the forward direction of the robot can be expressed as: wherein, is the total mass of the robot, is the acceleration of the robot in the forward direction, is the key state variable that the model needs to solve, is the traction force generated by the track, determined by the motor torque and the selected ground adhesion coefficient, is the rolling resistance of the track on a specific terrain, calculated by the selected soil parameter or rolling resistance coefficient, is the fluid resistance experienced by the robot in water, proportional to the fluid density, the square of the robot speed, and the hydrodynamic drag coefficient, is the thrust generated by the propeller, whose magnitude is determined by a nonlinear function or a lookup table to accurately simulate the actual working conditions.

[0039] It should be noted that the ground adhesion coefficient is calculated by the motor torque and ground pressure, or by looking up the terrain-coefficient table, the soil parameter is back calculated by the track settlement model or by looking up the table, and the hydrodynamic drag coefficient is calculated by stages according to the water depth. These are well-known parameter determination methods for those skilled in the art and can be directly implemented.

[0040] The model output data for subsequent prediction is generated by numerically solving the dynamic equations.

[0041] It should be noted that since the dynamic equations are ordinary differential equations containing nonlinear terms, it is difficult to obtain an analytical solution, therefore, a numerical method such as the fourth-order Runge-Kutta method is used to iteratively solve the dynamic equations. At each discrete time step, the acceleration, speed, and position at the next time instant are calculated based on the current control input, i.e., the motor speed, and the robot state, i.e., the speed and position. After solving for a continuous number of time steps, the motion state trajectory of the robot for a small period of time in the future is obtained. This trajectory is the model output data, which provides accurate prediction of the future behavior of the system for the subsequent prediction control module.

[0042] This method establishes a dynamically adaptive coupled dynamic model that converts abstract terrain recognition results and mode demand signals into accurate physical model parameters, and further constructs a mathematical model that can accurately describe and predict the complex dynamic behavior of the robot. This not only provides forward-looking insight for the system, but also quantifies the interaction between the track and the propeller in mixed working conditions, converting the complex coupling problem into a solvable mathematical problem, laying an important theoretical and data foundation for achieving precise and smooth decoupling control.

[0043] Specifically, the process of predicting the coupling influence degree in the future time period by using the rolling horizon prediction method is a key link for realizing the forward-looking control by using the dynamic model established in the previous step. The process first initializes the prediction horizon N and the control horizon M, which define the length of the forward prediction of the model and the length of the control sequence generated by the controller in one optimization, and the prediction horizon N is usually greater than the control horizon M to ensure sufficient predictability of the system to future changes.

[0044] In specific embodiments of the application, the process of predicting the coupling influence degree in the future time period by using the rolling horizon prediction method includes: in each prediction step, calculating the interaction force between the track and the propeller based on the coupling dynamic model to generate interaction force data.

[0045] Specifically, the track-propeller-vehicle body coupling dynamic model obtained in the previous link is used for open-loop simulation, which means that based on the currently measured robot state, the system simulates the motion trajectory of the robot under the action of a future control sequence, and in this simulation process, the model calculates the interaction force between the track system and the propeller system at each future time to generate interaction force data. This interaction force is particularly significant in the water-land transition area, for example, when the track moves in the mud, it generates both traction on the ground and thrust on the water, and the water flow also generates resistance on the track. These complex mechanical effects are all included in the interaction force data.

[0046] According to the interaction force data, the coupling influence degree is quantitatively evaluated by a cost function, and the specific evaluation formula is: .

[0047] wherein, is the total coupling influence degree quantization value in the future prediction horizon, and represent the speed components independently generated by the track and the propeller, respectively, and the square term of the speed difference between the two represents the conflict or redundancy of the two in the propulsion effect, i.e., the motion coupling, and are the acceleration components caused by the track and the propeller system, respectively, and the product term of the two represents the degree of vibration coupling, which is negative when the two vibrate in opposite directions, indicating the existence of harmful vibration, and are preset weight coefficients for adjusting the proportion of different coupling forms in the total cost, and the speed and acceleration components can be obtained from the intermediate calculation results of the coupling dynamic model, and the kth time to the k+N-1th time represents the future prediction horizon.

[0048] It is to be noted that the preset weight coefficient is specifically preset in an experience weight range according to a motion mode, and then is optimized by using a gradient descent method combined with a motion deviation, and the weight setting and adjustment logic can be reproduced by a person skilled in the art according to conventional experience, without additional special description.

[0049] The coupling influence degree after the quantification evaluation is updated to the prediction result.

[0050] Specifically, the calculated J_couple is taken as part of the prediction result, and the prediction result not only includes the prediction of the future state, but more importantly, includes the quantification evaluation of the coupling effect possibly caused by the future control strategy. In a rolling time domain framework, the controller seeks an optimal control sequence to minimize the total cost function including J_couple, and then applies the first element of the optimal control sequence to the system, and repeats the whole process in the next control period. By continuously predicting forward, evaluating the coupling influence and optimizing the control input, the system can actively suppress the coupling effect.

[0051] The method changes the control decision from passive response to active prediction by introducing a rolling time domain prediction mechanism. It not only predicts the motion state of the robot, but also quantifies and predicts the complex mutual influence between the track and the propeller, two power systems. This forward-looking evaluation of the degree of future coupling influence enables the control system to plan the optimal power distribution strategy in advance, fundamentally avoiding or weakening the occurrence of adverse coupling, thereby achieving significant improvement in the smoothness, stability and energy efficiency of the control.

[0052] Step four, based on the prediction result, dynamically adjusting the power distribution between the track motor and the propeller through an adaptive decoupling controller to generate power distribution parameters.

[0053] Specifically, the power distribution between the track motor and the propeller is dynamically adjusted by the adaptive decoupling controller in the present application, which is the core execution link of the whole control strategy. The process first inputs the prediction result generated in the previous step, especially the quantified coupling influence degree J_couple and the predicted motion state deviation, into a fuzzy logic controller as input, and outputs a preliminary control parameter, which is a control instruction of adjustment direction and approximate amplitude. It reflects the macro-strategic adjustment of the complex and nonlinear coupling problem, and the predicted motion state deviation specifically refers to the deviation between the predicted motion state of the robot and the expected motion state.

[0054] In a specific embodiment of the present application, the dynamic adjustment of the power distribution by the adaptive decoupling controller includes: inputting the prediction result into a fuzzy logic controller to generate a preliminary control parameter.

[0055] The fuzzy logic controller contains a series of "IF-THEN" rule base set based on expert experience, such as "IF coupling influence degree is large AND motion deviation is small THEN significantly increase decoupling degree", which converts precise numerical input into fuzzy sets through three steps of fuzzification, reasoning and defuzzification, and finally outputs a preliminary control parameter according to the rule base.

[0056] A proportional-integral-derivative controller is used to fine-tune the preliminary control parameter to generate an adjusted control parameter.

[0057] Specifically, in order to improve the dynamic response speed and steady-state accuracy of control, a parallel proportional-integral-derivative (PID) controller is used to fine-tune the preliminary control parameter. The PID controller calculates the proportional, integral and derivative components according to the error between the actual motion state and the desired state of the robot, and superimposes them with the output of the fuzzy logic controller. This forms a composite control structure, in which fuzzy control is responsible for dealing with the nonlinearity and uncertainty of the system, and PID control is responsible for quickly eliminating errors and ensuring the stability of the system. After fine-tuning by the PID, a more accurate adjusted control parameter is generated. The composite control law can be expressed as: wherein, is the final adjusted control parameter generated, is the preliminary control parameter output by the fuzzy logic controller, is the motion state error at the current time, i.e. the difference between the desired value and the actual measured value, and are the integral and derivative of the error, respectively, , , , are the gain coefficients of the fuzzy controller and the PID components, respectively, which are adaptively adjusted according to the operating conditions.

[0058] According to the adjusted control parameter, the power distribution ratio between the track motor and the propeller is calculated to generate the power distribution parameter.

[0059] For example, if U_final indicates that forward power needs to be increased while coupling vibration needs to be reduced, the power distribution module will decide to allocate 70% of the total power to the track and 30% to the propeller, and apply a specific phase difference to the control signals of the two, which directly determines the specific instructions sent to the underlying driver at the next time, and is the final basis for the execution of the decoupling control strategy.

[0060] The method realizes the organic unification of macroscopic strategy and microcosmic fine adjustment of the complex coupling system by constructing a compound controller combining fuzzy logic and PID. The introduction of the fuzzy logic controller enables it to handle the nonlinearities and uncertainties that the prediction model cannot completely cover, and the parallel use of the PID controller guarantees the rapidity and robustness of the control. This structure enables the adaptive decoupling controller not only to actively adjust according to the prospective prediction results, but also to respond quickly to real-time feedback errors, ultimately generating intelligent and accurate power distribution parameters, ensuring that the robot can achieve efficient and stable decoupling operation under various complex working conditions.

[0061] Specifically, the further optimization of the dynamic adjustment of the power distribution process in the application is realized by integrating multiple advanced control strategies, which work with the main controller to cope with specific challenges.

[0062] In a specific embodiment of the application, the dynamic adjustment of the power distribution further includes: when a mode demand signal of land-water transition is identified, activating a soft switching mechanism to smooth the change of the power distribution parameters.

[0063] The soft switching mechanism controls the change curve of the track power and the propeller thrust by designing a smooth transition function, such as a Sigmoid function, instead of a simple linear increase or decrease or step switching. Instead, it gradually and nonlinearly reduces the power output of the track motor and smoothly increases the thrust of the propeller according to the transition progress, such as the percentage of the vehicle body entering the water determined by the water depth sensor data. This gradual power handover avoids the impact and attitude instability caused by sudden switching of the power source, ensuring the motion continuity and attitude stability of the robot during the process of entering or leaving the water area.

[0064] The phase compensation algorithm adjusts the phase of the propeller control signal to offset the vibration coupling generated by the track movement.

[0065] The phase compensation algorithm monitors the vibration frequency and phase of the track and propeller in real time by performing a fast Fourier transform (FFT) on the motor current sensor data. When a specific frequency coupling vibration is detected between the two, a compensation phase angle is calculated, which is then applied to the control signal of the propeller motor. By fine-tuning the timing of the propeller thrust, it generates a force opposite to the direction of the track vibration, thereby actively offsetting or weakening the coupling vibration. The compensation process is closed-loop, capable of continuously tracking and suppressing dynamically changing vibrations.

[0066] It should be noted that the vibration frequency phase is analyzed by FFT, the compensation phase angle is calculated, and the vibration cancellation related means such as the threshold closed-loop adjustment control signal, which belongs to the conventional method of vibration control in the prior art, that is, the vibration suppression of industrial equipment, the coupling control of the vehicle power system and the like have been widely applied in such technology, and the skilled person can directly reproduce the vibration cancellation function based on the prior art knowledge and the basic conditions of the motor current sensor, the propeller control signal adjustment and the like in the present application.

[0067] According to the real-time energy efficiency and motion stability indicators, the power distribution parameters are optimized online.

[0068] Specifically, the real-time energy efficiency of the robot is continuously monitored, which is evaluated by the ratio of total input power to actual output speed, and the motion stability indicator is evaluated by the acceleration variance in the inertial measurement unit data. These performance indicators are associated with the power distribution parameters generated in the previous link, and an online optimization algorithm such as gradient descent or particle swarm optimization algorithm is used to continuously fine-tune the key parameters of the main controller such as fuzzy logic and PID controller in the background. This means that even if the preset control rules are self-learned and evolved according to the actual performance of the robot, the entire power distribution strategy can be adapted to new environments or changes in the performance of the robot, achieving long-term optimal operation.

[0069] The method deepens and expands the basic power distribution strategy from multiple dimensions by integrating soft switching mechanism, phase compensation algorithm and real-time optimization mechanism. It not only distributes power, but also carefully arranges the synergistic effect of power in time, phase and efficiency. Soft switching solves the smoothness problem of mode transition, phase compensation solves the vibration coupling problem in dynamic operation, and real-time optimization ensures the long-term adaptability and optimality of the system. The synergistic effect of these strategies makes the amphibious capability of the robot not a simple function superposition, but a performance fusion and sublimation at the system level, ultimately showing excellent motion stability, high energy efficiency and super strong adaptability to complex environments.

[0070] Step five, according to the power distribution parameters, outputting track motor control signal and propeller control signal to realize the decoupling of motion coupling.

[0071] Specifically, the process of outputting the track motor control signal and the propeller control signal in the application is the final execution exit of the entire decoupling control algorithm chain, which converts the abstract power distribution parameters calculated by the upper controller into specific instructions that can be recognized and executed by the bottom hardware driver. The process first needs to convert the power distribution parameters, which usually exist in the form of percentages or dimensionless coefficients, into motor drive instructions with clear physical meaning. This conversion process relies on a pre-calibrated motor performance model that describes the relationship between the input signals of the motor, such as PWM duty cycle or CAN bus message, and its actual output, such as speed and torque.

[0072] In specific embodiments of the application, the output track motor control signal and propeller control signal include converting the power distribution parameters combined with total speed and steering instructions into target drive instructions for track motors and propeller motors.

[0073] The track motor control signal is generated by a bottom closed-loop controller according to the target drive instruction.

[0074] The propeller control signal is generated by the bottom closed-loop controller according to the target drive instruction.

[0075] For example, the part of the power distribution parameter allocated to the track is combined with the desired total speed and steering instruction of the robot to calculate the target speed of the left and right track motors, respectively. For example, if the power distribution parameter indicates that the track system should bear 70% of the total propulsion force, and the robot needs to move straight at a speed v, the target speed of the left and right track motors is It can be determined by the following formula: wherein, is the total demand thrust calculated according to the desired speed v, is the torque constant of the motor, and if there is a steering instruction, a difference will be added to the target speed of the two motors to calculate the target speed. After calculating the target speed, a bottom speed closed-loop controller, which is usually a PI controller integrated in the motor driver, generates a track motor control signal that can drive the motor to reach the target speed, such as a PWM waveform signal with a specific duty cycle or a CAN bus message that meets a specific protocol.

[0076] The operation of generating the thruster control signal is similar to the track motor, and the part of the power distribution parameter allocated to the thruster is calculated to obtain the target thrust or rotating speed of the thruster motor. The calculation also needs to consider the hydrodynamic characteristic model of the thruster, which describes the nonlinear relationship between the rotating speed and the generated thrust. After the target rotating speed is obtained, the corresponding thruster control signal is generated through a speed or current closed-loop controller, and is sent to the thruster motor driver to accurately control the thrust output of the thruster.

[0077] The whole process ensures that the intention of the upper decoupling control strategy can be accurately transmitted to the physical execution layer of the robot, not only completes the conversion from the abstract parameter to the specific signal, but also ensures the accurate execution of the instruction through the closed-loop control of the bottom layer, and compensates the influence of the actual factors such as motor load change and power voltage fluctuation.

[0078] The method successfully converts the complex and deeply optimized power distribution strategy into accurate and reliable physical driving of the track and thruster execution mechanisms through a clear and model-based instruction conversion and bottom-layer closed-loop control process. It constitutes a key bridge connecting intelligent decision-making and physical action, and is a necessary link to ensure that the entire decoupling control system can effectively act on the robot entity and finally exhibit excellent performance in the real world. The stability and accuracy of this process directly determine whether the theoretical advantages brought by the upper algorithm can be converted into actual performance improvement.

[0079] Specifically, the real-time optimization and learning steps included in the present application give the entire decoupling control system the ability to continuously evolve, making it go beyond fixed parameter settings and adapt to long-term changes.

[0080] In specific embodiments of the present application, the method further comprises a real-time optimization and learning step: Collecting historical operation data including the environment state data, the power distribution parameters and system performance indicators.

[0081] Specifically, the historical operation data and current environment feedback data are continuously collected, and key data points under different terrains and different motion modes are recorded, including sensor inputs, intermediate decision variables of the controller such as power distribution parameters, actual outputs of the actuator such as motor current and rotating speed, and performance evaluation indicators of the system such as energy consumption, vibration amplitude, task completion time, etc. These data are structuredly stored in an on-board database to form a growing experience knowledge base.

[0082] Using an online optimization algorithm, the internal parameters of the adaptive decoupling controller are adjusted according to the historical operation data.

[0083] Specifically, periodically or triggered by certain events, such as detecting a persistent decline in system performance, an online optimization algorithm is used to adjust internal parameters, mainly referring to the rule weights in the aforementioned fuzzy logic controller, the membership function shape, and the gain coefficients in the PID controller. For example, a model-based reinforcement learning method is used to treat the decoupling control process as a Markov decision process. This method evaluates the long-term return, i.e., the comprehensive performance indicator, brought by different internal parameters in a specific state by analyzing historical operation data. The goal is to learn an optimal strategy π that can map the system state to the optimal internal parameter set θ. The optimization process can be represented as: ; where, and are the updated and pre-updated internal parameter sets, is the learning rate, which controls the step size of internal parameter updates, is the gradient of the performance indicator J with respect to the internal parameters θ, indicating the direction of internal parameter adjustment. The gradient can be estimated by analyzing historical data using algorithms such as policy gradient. By continuously fine-tuning internal parameters along the gradient direction that can improve system performance, the control strategy is continuously optimized.

[0084] Through a machine learning model, the mapping relationship between the environment and the control strategy is learned according to the historical operation data to improve the generalization ability of the adaptive decoupling controller.

[0085] Specifically, the goal of the machine learning model is to learn and predict the optimal dynamic model parameters or control strategy parameters under specific environmental characteristics and vehicle state. For example, when the robot enters a new type of muddy terrain that has never been seen before, even if there is no corresponding preset parameter in the parameter library, the machine learning model can infer a set of most suitable model parameters based on the current sensor readings. This ability enables the control system to quickly adapt to new environments, reducing the dependence on manual pre-labeling and parameter tuning, greatly improving the autonomy and environmental adaptability of the robot.

[0086] This method transforms the decoupling control system from a static, rule-based system to a dynamic, self-improving intelligent system by introducing a closed-loop learning and optimization mechanism. Through continuous data accumulation, online parameter optimization, and the generalization ability of machine learning models, the robot can become more efficient, more stable, and more adaptable to complex and changing environments in the long run. Not only does it solve the problem of parameter tuning at the initial deployment, but more importantly, it gives the robot the core ability to cope with unknown challenges and achieve long-term performance evolution, truly realizing adaptive and intelligent control strategies.

[0087] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, the above-described embodiments can be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art can be aware that the algorithm steps of the examples described in connection with the embodiments disclosed in the specification can be implemented by an electronic hardware, or a combination of computer software and electronic hardware. Whether the functions are performed by hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0088] In addition, each functional module in each embodiment of the present application can be integrated into one processing module, or each module can exist physically independently, or two or more modules can be integrated into one module.

[0089] The above is merely specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered in the protection scope of the present application, therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0090] Finally, the above is merely an example and description of the present application, any person skilled in the art can make various modifications or supplements to the described specific embodiments or use similar ways to replace, as long as it does not deviate from the concept of the present application or exceed the scope defined by the present application, which should be covered in the protection scope of the present application.

Claims

1. A decoupling control method for the motion coupling between the track and the propeller, characterized in that, include: Step 1: Acquire multi-source sensor data and perform fusion processing on the multi-source sensor data to generate environmental state data and vehicle body state data; Step 2: Based on the environmental state data and the vehicle state data, identify the current terrain type and motion mode requirements, and generate terrain recognition results and mode requirement signals; Step 3: Based on the terrain recognition results and the pattern requirement signal, construct a track-propeller-vehicle coupled dynamics model, and use the rolling time domain prediction method to predict the degree of coupling influence in future time periods, and generate prediction results; Step 4: Based on the prediction results, dynamically adjust the power distribution between the track motor and the propeller using an adaptive decoupling controller to generate power distribution parameters; Step 5: Based on the power distribution parameters, output the track motor control signal and the propeller control signal to achieve decoupling of motion coupling.

2. The decoupling control method for the motion coupling between the track and the propeller according to claim 1, characterized in that, The process of acquiring multi-source sensor data and generating environmental state data and vehicle state data includes: Data from water depth sensors, terrain recognition cameras, inertial measurement units, and motor current sensors are acquired to form a raw sensor data set. The water depth sensor measures the distance between the vehicle body and the bottom of the water or the ground; the terrain recognition camera data is a series of image frames reflecting the terrain ahead; the inertial measurement unit data is the raw readings of the three-axis acceleration and three-axis angular velocity; and the motor current sensor data reflects the working load of the track drive motor and the propeller motor. The original sensor data set is filtered and corrected to generate preprocessed sensor data; The preprocessed sensor data is input into a data fusion algorithm, which outputs the environmental state data and the vehicle body state data.

3. The decoupling control method for the motion coupling between the track and the propeller according to claim 1, characterized in that, The requirement to identify the current terrain type and movement pattern includes: Water depth features and terrain texture features are extracted from the environmental state data to generate an environmental feature vector; From the vehicle body state data, attitude features and motion speed features are extracted to generate a vehicle body feature vector; The environmental feature vector and the vehicle body feature vector are input into the classification model, and the terrain recognition result and the pattern requirement signal are output. The classification process can be represented by the following function: in, This represents the output terrain recognition results and pattern requirement signals. This represents the trained classification model function. This represents the environmental feature vector. This represents the vehicle body feature vector. This indicates that two feature vectors are concatenated.

4. The decoupling control method for the motion coupling between the track and the propeller according to claim 1, characterized in that, The construction of the track-propeller-vehicle coupled dynamics model includes: Based on the terrain recognition results and the pattern requirement signal, a set of model parameters is matched and selected from the parameter library; Based on the model parameters, a dynamic equation is established that includes fluid resistance, track ground pressure, and thruster nonlinearity. The simplified dynamic equation along the robot's forward direction can be expressed as: in, It is the total mass of the robot. It is the robot's acceleration along the forward direction, and it is a key state variable that the model needs to solve for. The traction force generated by the tracks is determined by both the motor torque and the selected ground adhesion coefficient. It is the rolling resistance of the track under specific terrain conditions, and its magnitude is calculated from the selected soil parameters or rolling resistance coefficient. This is the fluid resistance experienced by the robot in water, which is proportional to the fluid density, the square of the robot's velocity, and the hydrodynamic drag coefficient. The thrust generated by the propeller is determined by a nonlinear function or a lookup table method to accurately simulate actual working conditions. By numerically solving the dynamic equations, model output data is generated for subsequent predictions.

5. The decoupling control method for the motion coupling between the track and the propeller according to claim 1, characterized in that, The method of predicting the degree of coupling influence in future time periods using the rolling time-domain prediction method includes: Within each prediction step, the interaction force between the track and the propeller is calculated based on the coupled dynamics model to generate interaction force data. Based on the interaction force data, the degree of coupling influence is quantitatively evaluated using a cost function, the specific evaluation formula of which is as follows: ; in, It is a quantitative value of the total coupling influence in the future prediction time domain. and These represent the velocity components independently generated by the tracks and the propeller, respectively. The squared term of the velocity difference between the two characterizes the conflict or redundancy in their propulsive action, i.e., motion coupling. and These are the acceleration components caused by the track and propeller systems, respectively. The product of these two components characterizes the degree of vibration coupling. When the two systems vibrate in opposite directions, this term is negative, indicating the presence of harmful vibration. and These are preset weighting coefficients used to adjust the proportion of different coupling forms in the total cost. The velocity and acceleration components can be obtained through intermediate calculation results of the coupled dynamics model. The time from the kth moment to the k+N-1th moment represents the future prediction time domain. The quantitatively assessed degree of coupling influence is then updated in the prediction results.

6. The decoupling control method for the motion coupling between the track and the propeller according to claim 1, characterized in that, The dynamic adjustment of power allocation via an adaptive decoupling controller includes: The prediction results are input into the fuzzy logic controller to generate preliminary control parameters; The initial control parameters are finely adjusted using a proportional-integral-derivative controller to generate the adjusted control parameters; Based on the adjusted control parameters, the power distribution ratio between the track motor and the propeller is calculated, and the power distribution parameters are generated.

7. The decoupling control method for the motion coupling between the track and the propeller according to claim 1, characterized in that, The dynamic power allocation adjustment also includes: Upon identifying a mode demand signal for land-water transition, a soft switching mechanism is activated to smooth changes in the power allocation parameters; The phase of the thruster control signal is adjusted by a phase compensation algorithm to cancel out the vibration coupling generated by the track movement; The power distribution parameters are optimized online based on real-time energy efficiency and motion stability indicators.

8. The decoupling control method for the motion coupling between the track and the propeller according to claim 1, characterized in that, The output track motor control signal and propeller control signal include: The power distribution parameters are combined with the total speed and steering commands to convert them into target drive commands for the track motors and propeller motors. The track motor control signal is generated by the underlying closed-loop controller according to the target drive command. The underlying closed-loop controller generates the thruster control signal based on the target drive command.

9. The decoupling control method for the motion coupling between the track and the propeller according to claim 1, characterized in that, It also includes real-time optimization and learning steps: Collect historical operating data, including the environmental status data, the power distribution parameters, and system performance indicators; The internal parameters of the adaptive decoupling controller are adjusted based on the historical operating data using an online optimization algorithm. The machine learning model learns the mapping relationship between the environment and the control strategy based on the historical operating data, thereby improving the generalization ability of the adaptive decoupling controller.

Images