A wheel dynamics-based attitude control method for wheeled amphibious vehicles

Abstract

The application discloses a kind of based on wheel dynamics amphibious vehicle posture control method, belong to amphibious vehicle technical field.The application is aimed at the posture control problem of wheeled amphibious vehicle, the influence of internal wheel lift-drag characteristics and external wave disturbance on vehicle body posture is considered in the dynamics model of vehicle, improve control precision, speed and robustness.The application can calculate the wheel lift-drag mechanics characteristics, vehicle speed in real time by fusing dynamics model and deep deterministic policy gradient method, using dynamics parameters;Deep deterministic policy gradient method can change vehicle body posture to controllable range by outputting wheel speed in a short time, so that vehicle body maintains stable sailing posture under high speed condition.The application introduces wheel dynamics model as external input in simulation environment, can output different speed control strategies under variable environmental input conditions by giving rewards and penalties to amphibious vehicle agent.

Description

Technical Field

[0001] This invention relates to a method for attitude control of a wheeled amphibious vehicle based on wheel dynamics, belonging to the field of amphibious vehicle technology. Background Technology

[0002] A wheeled amphibious vehicle is a special vehicle that uses the rotation of its wheels for propulsion and can travel on both land and water. Because it does not require an additional propeller or waterjet propulsion system when navigating water, it offers advantages over traditional amphibious vehicles, including smaller size, simpler structure, lower cost, and easier maintenance. It has broad applications in industries such as river monitoring, aquaculture, emergency rescue, and recreational photography. The wheel structure of a wheeled amphibious vehicle is often specially designed to generate forward propulsion and upward lifting force when propelled by water. The magnitude of these forces depends on the depth of immersion and the rotational speed of the wheels. Increased lifting force causes the vehicle to float, reducing its volume in the water, thus reducing drag and increasing speed. Each wheel is independently connected to a motor, directly controlled by the rotational speed signal from the central control unit, providing different propulsion and lifting forces to achieve various postures.

[0003] Wheeled amphibious vehicles are typically subjected to the combined effects of turbulence and currents in wave-like environments, making the development of high-speed stability control strategies crucial. However, despite the relatively simple structure of a wheeled amphibious vehicle with only one propulsion system, establishing its control system remains challenging. First, while amphibious vehicles operate with six degrees of freedom in water, they only have four wheel speed control inputs, making them a typical underactuated nonlinear system. Second, the wheels exhibit poor stability at high speeds and are highly susceptible to environmental disturbances, often resulting in significant pitching and a severe shift in the center of gravity. Third, compared to ships, amphibious vehicles are inherently weaker in terms of stability, wave resistance, and maneuverability, making them more sensitive to environmental disturbances such as waves and typhoons. In conclusion, achieving efficient and robust attitude control for high-speed wheeled amphibious unmanned vehicles in aquatic environments is a highly challenging task.

[0004] Currently, the attitude control method for amphibious vehicles mostly adopts the PID (proportion integration differentiation) control method. This is because the PID algorithm has a simple structure, resulting in a relatively complete design and control flow, which is easily mastered by engineers. However, the proportional, integral, and derivative units of the PID control method are simply linear combinations, making it difficult to handle nonlinear, strongly coupled multi-input systems. Furthermore, in complex, variable, and unpredictable amphibious operating environments with strong interference such as sea winds and waves, the relatively slow PID control strategy cannot guarantee the operational safety of amphibious vehicles. In addition, during high-speed navigation, the attitude of wheeled amphibious vehicles is greatly affected by the wheels. The control strategies for wheeled amphibious vehicles often ignore the influence of wheel forces on the vehicle body, leading to reduced control performance and consequently reduced tracking accuracy. Summary of the Invention

[0005] To address some or all of the technical problems existing in the prior art, this invention proposes an attitude control method for wheeled amphibious vehicles based on wheel dynamics. It employs a Deep Deterministic Policy Gradient (DDPG) algorithm, introducing a wheel dynamics model as an external input into the simulation environment. By providing reward and penalty stimuli to the amphibious vehicle agent, it can output different speed control strategies under varying environmental inputs. This method is particularly suitable for long-term active feedback control of wheeled amphibious vehicles where the influence of the external environment is unclear, enabling the amphibious vehicle to quickly enter a high-speed and stable navigation state under uncertainties such as model errors, environmental disturbances, and observation errors.

[0006] The objective of this invention is achieved through the following technical solution.

[0007] This invention discloses a method for attitude control of a wheeled amphibious vehicle based on wheel dynamics, comprising the following steps:

[0008] S1. Construct a simulation environment for the water surface motion of a wheeled amphibious vehicle, including a six-DOF maneuvering, fully coupled wave-resistant environment dynamic model for the wheeled amphibious unmanned vehicle, considering both wheel dynamics and external wave interference, as shown below:

[0009]

[0010] In the formula, m is the mass of the vehicle body; I xx ,I yy I zzThese represent the moments of inertia about the x, y, and z axes, respectively; subscript H represents the vehicle body force, which is related to the vehicle body structure; subscript P represents the propeller wheel force, determined from propeller wheel open-water test data; 1W and 2W represent the first-order wave force and second-order wave drift force, respectively, calculated based on the linear theory assumption under infinite water depth conditions, and are related to environmental factors; X, Y, and Z represent the sway, roll, and heave forces of the vehicle body, respectively; K, M, and N represent the roll, pitch, and yaw moments of the vehicle body, respectively; u, v, and w,

[0011] These represent the linear velocities and linear accelerations of the vehicle body in three directions, respectively; p, q, r, These are the Euler angles and Euler angle accelerations in three directions, respectively.

[0012] S2. Construct the amphibious vehicle intelligent agent, including the research object, initial state, system input, system output and termination conditions, establish the motion controller strategy, and determine the sampling time, simulation time and reward information function;

[0013] S3. Construct the Critic target network and Actor target network using the DDPG algorithm, and perform multiple rounds of training and update the agent in the simulation environment of S1.

[0014] S4. Test the control index data of the training results. If the control index does not converge, return to S3 for retraining. If the control index converges, encapsulate the final agent to select the optimal control output based on the input state information.

[0015] S5. In a real outdoor environment, test and deploy a control agent for wheeled amphibious vehicles. Through the trained agent, actively control the wheeled amphibious vehicles to enable them to quickly enter a high-speed and stable navigation state under the interference of model errors, environmental disturbances, and observation error uncertainties.

[0016] As a further improvement of the present invention, S2 clarifies the control strategy and method of the amphibious vehicle, wherein the specific steps are as follows, including S2-1 to S2-10;

[0017] S2-1. The research object is determined to be an amphibious unmanned vehicle that considers the force of wheel action.

[0018] S2-2. Determine the initial state of the vehicle, including the depth of the wheels submerged in water, wheel speed, position, and speed.

[0019] S2-3, The system input information is the immersion depth H of the four wheels. i T ;

[0020] S2-4, The system output information is the rotational speed N of the four wheels. i T ;

[0021] S2-5, The motion controller strategy is to ensure that the amphibious unmanned vehicle can navigate as stably and quickly as possible within the vehicle height range (-h, h);

[0022] S2-6, The termination condition is the end of the vehicle simulation time or the vehicle reaches the designated location;

[0023] S2-7. Determine the length of the time step for the interaction between the strategy and the environment, i.e., the sampling time T. s ;

[0024] S2-8. Determine the maximum simulation time T f ;

[0025] S2-9, The termination condition is the end of the vehicle simulation time or the vehicle reaches the designated location;

[0026] S2-10, where T is the time step in the strategy and environment. s The reward value is determined in the middle, and distributed to the robot's reward information function R in units of each time step. t ;

[0027] As a further improvement of the present invention, S3 trains and updates the amphibious vehicle agent in the simulation environment, wherein the specific steps are as follows, including S3-1 to S3-5;

[0028] S3-1. Construct the Critic target network and the Actor target network;

[0029] S3-2. Define the wheel immersion depth H, rotational speed N, vehicle position information P, and vehicle speed V as a set of states S, where state S is defined at a certain moment. t =[H t N t ,P t V t ];

[0030] S3-3. Select action N based on the current state. t+1 By interacting with the external environment, it generates the state information for the next moment and the reward function information for this moment;

[0031] S3-4. Determine whether to terminate the state update, and set the quintuple {S} t A t ,R t ,S t+1 ,EndOrNot} is stored in the experience replay set D;

[0032] S3-5 uses n samples from set D to calculate the current target feedback Q value and updates the parameters of the Critic target network and the Actor target-related network.

[0033] As a further improvement of the present invention, S4 will select the optimal output based on the motion controller strategy and the amphibious vehicle's state information, wherein the specific steps are as follows, including S4-1 and S4-2:

[0034] S4-1. In the simulation environment of S1, the control index data of the training results are used for testing to obtain the vertical position index of the vehicle body in the preset trajectory information.

[0035] S4-2. Determine whether the control index has converged within the vehicle height range (-h, h). If the index has not converged, return to S3 for retraining; if the index has converged successfully, save the target network.

[0036] As a further improvement of the present invention, S5 deploys the control strategy obtained in S4, wherein the specific steps are as follows, including S5-1 to S5-3:

[0037] S5-1. Real-time data on wheel immersion depth H is collected through the vehicle body IMU inertial measurement unit, wheel speed signal N is collected through the motor, position information P is collected through Beidou positioning information, and vehicle speed information V is collected through the speed sensor.

[0038] S5-2. Using wave measurement equipment for ocean surface directional waves and surface currents, obtain directional wave patterns and surface current wave information, and further obtain first-order wave force and second-order wave drift force.

[0039] S5-3, the system state variable S at time t t =[H t N t ,P t V t The information is input into the vehicle body controller to obtain the optimal control information S of the rotational speed of the four wheels of the system at the next moment. This information is then used for active feedback control of the wheeled amphibious vehicle, enabling the amphibious vehicle to quickly enter a high-speed and stable navigation state under the interference of model errors, environmental disturbances, and observation error uncertainties.

[0040] Beneficial effects:

[0041] 1) The present invention discloses a wheeled amphibious vehicle attitude control method based on wheel dynamics. In view of the attitude control problem of wheeled amphibious vehicles, the method considers the internal wheel lift and drag characteristics and the influence of external wave disturbance on the vehicle attitude in the vehicle dynamics model, thereby improving the control accuracy and speed and enhancing the robustness of the control system.

[0042] 2) The present invention discloses a wheeled amphibious vehicle attitude control method based on wheel dynamics. By integrating the dynamic model and the deep deterministic policy gradient method, the dynamic parameters can be used to calculate the wheel lift and drag mechanical characteristics and vehicle speed in real time. The deep deterministic policy gradient method can change the vehicle attitude to a controllable range in a short time by outputting the wheel speed, so that the vehicle can maintain a stable navigation attitude at high speed. Attached Figure Description

[0043] Figure 1 This is a diagram showing the forces acting on the wheel.

[0044] Figure 2 This is a schematic diagram of the control system for a wheeled amphibious vehicle.

[0045] Figure 3 This is a flowchart of the control process for wheeled amphibious vehicles. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0047] like Figure 1 The diagram shows the forces acting on the wheels of a four-wheel-drive amphibious vehicle. The surfaces of the wheels are specially designed to generate upward lift and forward propulsion when rotating in water. By changing the rotational speed of each wheel, different dynamic characteristics can be exhibited, thereby achieving attitude and position control of the vehicle in the water. The vehicle's motion can generally be categorized into seven states: hovering, straight-ahead motion, reversing motion, turning left forward, turning right forward, turning left rearward, and turning right rearward.

[0048] This example presents an attitude control method for a wheeled amphibious vehicle that considers wheel dynamics and external wave disturbances. The specific process is as follows: Figure 2 As shown, the specific implementation method of the attitude control method for wheeled amphibious vehicles that considers wheel dynamics and external wave interference is as follows:

[0049] S1. Construct a simulation environment for the water surface motion of a wheeled amphibious vehicle. The dynamic model of the wheeled amphibious unmanned vehicle's six-DOF maneuvering, wave-resistant, fully coupled environment considers both wheel dynamics and external wave interference, and is expressed as follows:

[0050]

[0051] In the formula, m is the mass of the hull; I xx ,I yy I zzThese represent the moments of inertia about the x, y, and z axes, respectively; the subscript H represents the vehicle body forces, which can be divided into inertial forces and viscous forces. The former is related to the vehicle's acceleration, while the latter is related to the characteristic parameters of the vehicle structure and is generally estimated using empirical formulas; the subscript P represents the paddlewheel forces, which can be divided into vertical lifting forces and horizontal propulsion forces. The former increases continuously with the depth of the wheel submerged in water and remains positive; the latter decreases continuously with the depth of the wheel submerged in water, reaching zero when the water level is at 1 / 2 of the wheel's height; the amplitudes of both increase with the rotational speed, and the specific hydrodynamic characteristics can be determined from the paddlewheel open-water test data; 1W and 2W represent the first-order wave force and the second-order wave drift force, respectively, calculated based on the linear theory assumption under infinite water depth conditions and are related to environmental factors; X, Y, and Z represent the sway, roll, and heave forces of the vehicle body, respectively; K, M, and N represent the roll, pitch, and yaw moments of the vehicle body, respectively; u, v, and w... These represent the linear velocities and linear accelerations of the vehicle body in three directions, respectively; p, q, r, These represent the Euler angles and Euler angular accelerations in three directions. This model primarily reflects the dynamic response of the propulsion and lifting forces provided by the four wheels of an amphibious vehicle across various coordinate axes under external environmental disturbances dominated by waves. The model's inputs are the disturbance quantities of the wave characteristics of the external wave environment and the wheel speed control quantities of the four wheels; the outputs are the motion state quantities of the wheeled amphibious vehicle (including three-dimensional spatial coordinates). Motion velocity component And the attitude angle E = [p, q, r]).

[0052] S2. Determine the amphibious vehicle's intelligent agent, including the research object, initial state, system input, system output, and termination conditions; establish a motion controller strategy; and determine the sampling time, simulation time, and reward information function.

[0053] S2-1. The research object is determined to be an amphibious unmanned vehicle considering the force of the wheels; a discrete motion state model of the vehicle body is constructed, and the motion state of the vehicle body is divided into seven states: hovering, straight-line motion, reversing motion, left front turning, right front turning, left rear turning and right rear turning;

[0054] S2-2, Determine the initial state S of the vehicle. t =[H t N t ,P t V t [Including the wheel immersion depth H] t Wheel speed N t Location information P t and sailing speed V t ;

[0055] S2-3, The system input is the immersion depth H of the four wheels.i T To determine the current attitude of the vehicle body;

[0056] S2-4, The system output information is the rotational speed N of the four wheels. i T ;

[0057] S2-5, the motion controller strategy is to ensure that the amphibious unmanned vehicle navigates as stably and quickly as possible within the vehicle's height range (-h, h), see appendix. Figure 3 ;

[0058] S2-6, The termination condition is the end of the vehicle simulation time or the vehicle reaches the designated location;

[0059] S2-7. Determine the length of the time step for the interaction between the strategy and the environment, i.e., the sampling time T. s =0.02s;

[0060] S2-8. Determine the maximum simulation time T f =50s, to reproduce the result, a fixed random number seed is used;

[0061] S2-9, The termination condition is the end of the vehicle simulation time or the vehicle reaches the designated location;

[0062] S2-10, where T is the time step in the strategy and environment. s The reward value is determined in the middle, and distributed to the robot's reward information function R in units of each time step. t The reward signal is set at each time step T of the policy and environment. s The reward value is determined in the middle, and the reward signal R is distributed to the robot in units of each time step. t :

[0063]

[0064] In the formula, α, β, γ, and μ are all positive values ​​and represent different reward signal weighting coefficients; The value represents the scalar speed of the amphibious vehicle; the larger the value, the stronger the reward signal. The term represents the total number of steps in the stage simulation time; a larger value indicates a longer simulation duration and a stronger reward signal. (zh) 2 The indicator reflects the vertical offset of the amphibious vehicle, with full surfacing as the baseline. The more it sinks, the stronger the penalty signal. The term represents the control quantity of the decision-maker. The larger the value, the more control quantity the system has, the less efficient the control system is, and the stronger the penalty signal generated.

[0065] S3. Using the DDPG algorithm, the Critic target network and Actor target network are constructed using the experience pool and dual network structure in DQN, so that the agent can be trained and updated in multiple rounds of continuous action space in the simulation environment of S1.

[0066] S3-1. Construct the Critic and Actor target networks. Use the Critic value function to define an input function containing the amphibious vehicle's state and action variables. These are processed through several layers of grid nodes and aggregated to generate the CriticOutput function. Create the Actor action function, which takes the environmental state information S as input. t =[H t N t ,P t V t Get the current estimated action of the network, randomize the estimated action and pass it into the Critic value function to get the maximum score;

[0067] S3-2. Define the wheel immersion depth H, rotational speed N, vehicle position information P, and vehicle speed V as a set of states S, where state S is defined at a certain moment. t =[H t N t ,P t V t ]; where the wheel immersion depth H i T =[h1,h2,h3,h4], Rotational speed N i T =[n1,n2,n3,n4], Vehicle position information P i T = [x,y,z] and vehicle speed V i T = [u,v,w] is defined as a set of states S, where the initial state S0 = [H0,N0,P0,V0]; the rotational speed N of the wheel is... i T =[n1,n2,n3,n4] represents the action quantities;

[0068] S3-3. Select action N based on the current state. t+1 By interacting with the external environment, it generates the state information for the next moment and the reward function information for this moment, and stores the function information on the server;

[0069] S3-4. Determine whether to terminate the state update based on whether the maximum simulation time has been reached or the vehicle has reached the preset state. If the amphibious vehicle has not reached the maximum simulation time and has not reached the designated location, then the quintuple set {S} will be updated. t A t ,R t ,St+1 ,EndOrNot} is stored in the experience replay set D, and the next iteration begins;

[0070] S3-5 uses n samples from set D to calculate the current target feedback Q value and updates the parameters of the Critic target network and the Actor target-related network.

[0071] S4. Test the control index data of the training results. If the control index does not converge, return to S3 for retraining. If the control index converges, encapsulate the final agent to select the optimal control output based on the input state information.

[0072] S4-1. In the simulation environment of S1, the control index data of the training results are used for testing to obtain the vertical position index of the vehicle body in the preset trajectory information.

[0073] S4-2. Determine whether the control index has converged within the vehicle height range (-h, h). If the index has not converged, return to S3 for retraining; if the index has converged successfully, save the target network.

[0074] S5. Test and deploy wheeled amphibious vehicle control agents in real-world outdoor environments.

[0075] S5-1. Real-time data on wheel immersion depth H is collected through the vehicle body IMU inertial measurement unit, wheel speed signal N is collected through the motor, position information P is collected through Beidou positioning information, and vehicle speed information V is collected through the speed sensor.

[0076] S5-2. Using wave measurement equipment for ocean surface directional waves and surface currents, obtain directional wave patterns and surface current wave information, and further obtain first-order wave force and second-order wave drift force.

[0077] S5-3, the system state variable S at time t t =[H t N t ,P t V t The information is input into the vehicle body controller to obtain the optimal control information S of the rotational speed of the four wheels of the system at the next moment.

[0078] The above detailed description further illustrates the purpose, technical solution, and beneficial effects of the invention. It should be understood that the invention is not limited to the scope of the specific embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.

Claims

1. A method for attitude control of a wheeled amphibious vehicle based on wheel dynamics, characterized in that: Includes the following steps, S1. Construct a simulation environment for the water surface motion of a wheeled amphibious vehicle. This includes a six-DOF (degree of freedom) maneuvering, wave-resistant, fully coupled dynamic model of the wheeled amphibious unmanned vehicle, considering both wheel dynamics and external wave interference. This model is represented as follows: In the formula, For vehicle body mass; , , These represent the moments of inertia about the x, y, and z axes, respectively; the subscript H represents the vehicle body force, which is related to the vehicle body structure; the subscript P represents the propeller wheel force, determined from propeller wheel open-water test data; 1W and 2W represent the first-order wave force and the second-order wave drift force, respectively, calculated based on the linear theory assumption under infinite water depth conditions, and are related to environmental factors; X, Y, and Z represent the sway, roll, and heave forces of the vehicle body, respectively; K, M, and N represent the roll, pitch, and yaw moments of the vehicle body, respectively. u, v, w , , These represent the linear velocities and linear accelerations of the vehicle body in three directions, respectively; p, q, r, , , These are the Euler angles and Euler angle accelerations in three directions, respectively. S2. Construct the amphibious vehicle intelligent agent, including the research object, initial state, system input, system output and termination conditions, establish the motion controller strategy, and determine the sampling time, simulation time and reward information function; S2 clarifies the control strategy and methods for amphibious vehicles, with the following specific steps, including S2-1 to S2-10; S2-1. The research object is determined to be an amphibious unmanned vehicle that considers the force of wheels. S2-2. Determine the initial state of the vehicle, including the depth of the wheels submerged in water, wheel speed, position, and speed. S2-3, The system input information is the immersion depth of the four wheels. ; S2-4, The system output information is the rotational speed of the four wheels. ; S2-5, The motion controller strategy is to ensure that the amphibious unmanned vehicle can navigate as stably and quickly as possible within the vehicle height range (-h, h); S2-6, The termination condition is the end of the vehicle simulation time or the vehicle reaches the designated location; S2-7. Determine the length of the time step for the interaction between the strategy and the environment, i.e., the sampling time. ; S2-8. Determine the maximum simulation time. ; S2-9, The termination condition is the end of the vehicle simulation time or the vehicle reaches the designated location; S2-10, for each time step in the strategy and environment. The reward value is determined in the middle, and distributed to the robot's reward information function in units of each time step. ; S3. Construct the Critic target network and Actor target network using the DDPG algorithm, and perform multiple rounds of training and update the agent in the simulation environment of S1. S4. Test the control index data of the training results. If the control index does not converge, return to S3 to retrain. If the control index converges, the final agent is encapsulated to select the optimal control output based on the input state information. S4 will select the optimal output based on the motion controller strategy and the amphibious vehicle's state information. The specific steps are as follows, including S4-1 and S4-2: S4-1. In the simulation environment of S1, the control index data of the training results are used for testing to obtain the vertical position index of the vehicle body in the preset trajectory information. S4-2. Determine whether the control index has converged within the vehicle height range (-h, h). If the index has not converged, return to S3 for retraining; if the index has converged successfully, save the target network. S5. In a real outdoor environment, test and deploy a wheeled amphibious vehicle control agent. Through the trained agent, actively control the wheeled amphibious vehicle to enable it to quickly enter a high-speed and stable navigation state under the interference of model error, environmental interference, and observation error uncertainty. S5 deploys the control strategy obtained in S4, and the specific steps are as follows, including S5-1 to S5-3: S5-1. Real-time data on wheel immersion depth H is collected through the vehicle body IMU inertial measurement unit, wheel speed signal N is collected through the motor, position information P is collected through Beidou positioning information, and vehicle speed information V is collected through the speed sensor. S5-2. Using wave measurement equipment for ocean surface directional waves and surface currents, obtain directional wave patterns and surface current wave information, and further obtain first-order wave force and second-order wave drift force. S5-3, The system state variables at time t The information is input into the vehicle body controller to obtain the optimal control information S of the rotational speed of the four wheels of the system at the next moment. This information is then used for active feedback control of the wheeled amphibious vehicle, enabling the amphibious vehicle to quickly enter a high-speed and stable navigation state under the interference of model errors, environmental disturbances, and observation error uncertainties.

2. The attitude control method for a wheeled amphibious vehicle based on wheel dynamics as described in claim 1, characterized in that: S3 trains and updates the amphibious vehicle agent in the simulation environment, and the specific steps are as follows, including S3-1 to S3-5; S3-1. Construct the Critic target network and the Actor target network; S3-2. Define the wheel immersion depth H, rotational speed N, vehicle position information P, and vehicle speed V as a set of states S, where at a certain moment the state is... ; S3-3. Select action N based on the current state. t+1 By interacting with the external environment, it generates the state information for the next moment and the reward function information for this moment; S3-4. Determine whether to terminate the state update and set the quintuple. Store in the experience replay set D; S3-5 uses n samples from set D to calculate the current target feedback Q value and updates the parameters of the Critic target network and the Actor target-related network.

Images