A collaborative control method for amphibious unmanned vehicle's water and land actuators

Abstract

The present invention relates to a method for coordinated control of amphibious actuators for an amphibious unmanned vehicle. This method, which belongs to the technical field of amphibious unmanned vehicles, addresses the existing problem of poor adaptability of amphibious unmanned vehicles in different scenarios. The method comprises the following steps: acquiring the amphibious unmanned vehicle's suspension state and attitude angle data at each moment; predicting the water model mismatch probability based on the current and historical suspension state and attitude angle data using a trained recurrent neural network; determining whether the current point is a key scene point based on the predicted water model mismatch probability; key scene points include landing point, departure point, entry point, and full float point; and calculating the actuator action instructions for the amphibious unmanned vehicle based on whether the current point is a key scene point and the predicted water model mismatch probability. This method achieves stable control of the amphibious unmanned vehicle in both water and land conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of amphibious unmanned vehicles, and in particular to a method for coordinated control of amphibious actuators of amphibious unmanned vehicles. Background Art

[0002] Amphibious unmanned vehicles (UAVs) are vehicles capable of traveling across land and water, and are therefore of significant research value. Because they carry two sets of relatively independent yet coupled actuators, one on land and one on water, and must navigate a variety of complex and distinct scenarios, such as on land, on water, and at the interface between land and water, they face the challenge of coordinating these two actuators for different scenarios. In both land and water conditions, both the land and water actuators must be controlled simultaneously. Existing technologies, however, mostly fail to consider the coordinated control of these two actuators and the stable switching between operating conditions, resulting in poor adaptability to diverse scenarios. Summary of the Invention

[0003] In view of the above analysis, an embodiment of the present invention aims to provide a method for collaborative control of amphibious actuators of an amphibious unmanned vehicle, so as to solve the problem of poor adaptability of existing amphibious unmanned vehicles to different scenarios.

[0004] In one aspect, an embodiment of the present invention provides a method for collaboratively controlling water and land actuators of an amphibious unmanned vehicle, comprising the following steps:

[0005] Obtain the suspension state data and attitude angle data of the amphibious unmanned vehicle at each moment;

[0006] At the current moment, the probability of water model mismatch is predicted based on the trained recurrent neural network according to the suspension state data and attitude angle data at the current moment and historical moments;

[0007] Determine whether the current scene is a key point based on the predicted water model mismatch probability; the key points of the scene include: landing point, water departure point, water entry point and full floating point;

[0008] The actuator action instructions of the amphibious unmanned vehicle are solved based on whether the current point is a key point in the scene and the predicted water model mismatch probability.

[0009] Based on the further improvement of the above method, whether the current point is a key point in the scene is determined based on the predicted water model mismatch probability, including:

[0010] If the water model mismatch probability predicted at the current moment is greater than 0, and the water model mismatch probability predicted at the previous moment is equal to 0, then the current moment is the landing point;

[0011] If the mismatch probability of the water model predicted at the current moment is greater than 90%, and the mismatch probability of the water model predicted at the previous moment is less than or equal to 90%, then the current point is the water departure point;

[0012] If the mismatch probability of the water model predicted at the current moment is less than 100%, and the mismatch probability of the water model predicted at the previous moment is equal to 100%, then the current moment is the entry point;

[0013] If the mismatch probability of the water model predicted at the current moment is less than 10%, and the mismatch probability of the water model predicted at the previous moment is greater than or equal to 10%, the current state is full floating point.

[0014] Based on the further improvement of the above method, the actuators of the amphibious unmanned vehicle include gear, throttle, water direction control mechanism and land direction control mechanism;

[0015] The actuator action instructions of the amphibious unmanned vehicle are calculated based on whether the current scene is a key point and the predicted water model mismatch probability, including:

[0016] Determine the current solution mode and gear instruction according to whether the current scene is a key point, wherein the solution mode includes water mode, land-water mode and land mode;

[0017] Calculate the throttle opening according to the current solution mode and the predicted water model mismatch probability;

[0018] If the current mode is land mode, the control instructions of the land direction control mechanism are determined according to the expected curvature; if the current mode is water mode, the control instructions of the water direction control mechanism are determined according to the expected curvature; if the current mode is land-water mode, the control instructions of the land direction control mechanism and the water direction control mechanism are determined according to the expected curvature.

[0019] Based on the further improvement of the above method, the current solution mode and gear instruction are determined according to whether the current scene is a key point, including:

[0020] If the current point is the water entry point, the solution mode is the water-land mode;

[0021] If the current mode is full floating point, the solution mode is water mode, and the gear command is the gear shift command;

[0022] If the current point is a landing point, the solution mode is the land-water mode, and the gear command is the gear engagement command;

[0023] If the current point is out of water, the solution mode is on land mode.

[0024] Based on the further improvement of the above method, the throttle opening is calculated according to the current solution mode and the predicted water model mismatch probability, including:

[0025] Determine the conversion coefficient between engine speed and unmanned vehicle speed based on the predicted water model mismatch probability;

[0026] If the current solution mode is the land mode or the water-land mode, the throttle opening is calculated according to the expected speed and the conversion coefficient;

[0027] If the current solution mode is the water mode, the expected speed is corrected, and the throttle opening is calculated based on the corrected expected speed and the conversion coefficient.

[0028] Based on the further improvement of the above method, the conversion coefficient between engine speed and unmanned vehicle speed is determined by the following formula based on the predicted water model mismatch probability:

[0029]

[0030] Where λ represents the conversion coefficient between engine speed and unmanned vehicle speed, represents the predicted water model mismatch probability, λ l represents the longitudinal kinematic model coefficient for onshore conditions, λ w is the longitudinal kinematic model coefficient for water working conditions.

[0031] Based on the further improvement of the above method, the expected speed is corrected, including:

[0032] Calculate the theoretical flow velocity characteristics and calculated flow velocity characteristics within the characteristic time period corresponding to the current moment;

[0033] Obtaining a correction coefficient based on a trained correction coefficient prediction model according to the theoretical flow velocity characteristics and the calculated flow velocity characteristics;

[0034] Correct the expected speed according to the correction factor.

[0035] Based on the further improvement of the above method, the recurrent neural network is trained in the following way:

[0036] The speed, engine speed, suspension state data, and attitude angle data of the amphibious unmanned vehicle are obtained at each moment in the entire history. The water model mismatch probability at each moment is calculated based on the speed and engine speed. The features of each moment are calculated based on the suspension state data and attitude angle data at each moment. The training sample set is constructed by taking the features of the k+1 moment as the input data of a sample and the average water model mismatch probability of the k+1 moments after the k+1 moment as the sample label.

[0037] Constructing a recurrent neural network model; training the constructed recurrent neural network model based on the training sample set to obtain a trained recurrent neural network model.

[0038] Based on the further improvement of the above method, the features of each moment are calculated based on the suspension state data and attitude angle data at each moment, including:

[0039] For each moment, the time period from the sth moment before the moment to the moment is the characteristic time period corresponding to the moment;

[0040] The maximum value of the suspension state data, the average value of the suspension state data, the variance of the suspension state data, the maximum value of the attitude angle data, the average value of the attitude angle data and the variance of the attitude angle data in the characteristic time period corresponding to the moment are calculated as the features corresponding to the moment.

[0041] Based on the further improvement of the above method, the water model mismatch probability at each moment is calculated according to the speed and engine speed in the following way:

[0042]

[0043] Where v represents speed, n represents engine speed, and λ w represents the kinematic model coefficient of the water working condition, λ l Represents the kinematic model coefficient of onshore conditions.

[0044] Compared with the existing technology, the present invention obtains the suspension status data and attitude angle data of the amphibious unmanned vehicle at each moment; predicts the water model mismatch probability at each current moment, determines whether the current moment is a scene key point based on the predicted water model mismatch probability, and confirms the solution mode according to whether the current moment is a scene key point and the predicted water model mismatch probability to solve the actuator action instructions of the amphibious unmanned vehicle, thereby realizing coordinated stable process control of the water actuator and the land actuator.

[0045] In the present invention, the above-mentioned technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of the present invention will be described in the following description, and some advantages will become apparent from the description or be learned through practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings are only used for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. Throughout the drawings, the same reference symbols denote the same components.

[0047] Figure 1 This is a flow chart of a method for collaborative control of water and land actuators of an amphibious unmanned vehicle according to an embodiment of the present invention;

[0048] Figure 2 Schematic diagram of key points of an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, and are not used to limit the scope of the present invention.

[0050] A specific embodiment of the present invention discloses a method for cooperatively controlling the water and land actuators of an amphibious unmanned vehicle. Figure 1 As shown, the following steps are included:

[0051] S1. Obtain the suspension state data and attitude angle data of the amphibious unmanned vehicle at each moment;

[0052] S2. At the current moment, predict the water model mismatch probability based on the trained recurrent neural network according to the suspension state data and attitude angle data at the current moment and historical moments;

[0053] S3. Determine whether the current scene is a key point based on the predicted water model mismatch probability; the key points of the scene include: landing point, water departure point, water entry point, and full floating point;

[0054] S4. Calculate the actuator action instructions of the amphibious unmanned vehicle based on whether the current point is a key point in the scene and the predicted water model mismatch probability.

[0055] The autonomous planning module of the amphibious unmanned vehicle uses intelligent sensors, such as image sensors, to perform visual detection to obtain environmental information. According to the mission objectives, status information and environmental information of the unmanned platform, it plans the driving strategy based on intelligent algorithms such as machine learning, reinforcement learning, and deep learning, and sends control instructions to the collaborative control end to control the unmanned platform to run towards the planned target. The working conditions of the amphibious unmanned vehicle include water conditions, land conditions and water and land conditions. The amphibious unmanned vehicle water and land actuator collaborative control method of the present invention is mainly applied to the water and land conditions of amphibious unmanned vehicles. During implementation, the suspension displacement change data is collected in real time by the suspension displacement sensor, and the attitude angle data is collected in real time by the combined navigation system, such as inertial navigation, so as to obtain the suspension state data and attitude angle data of the amphibious unmanned vehicle at each moment.

[0056] The water model mismatch probability is used to characterize the response state of the actuator of the amphibious unmanned vehicle under water and land conditions, and is an important feedback for the vehicle to perceive the characteristics of the water-land interface scene. Figure 2 As shown in Figure 1, as the amphibious unmanned vehicle enters the water, its tracks are affected by the water environment, causing the water model mismatch probability to gradually decrease. During the landing process, the tracks gradually touch the ground, causing the water model mismatch probability to gradually increase. The water model mismatch probability predicted in step S2 is the predicted water model mismatch probability for the future.

[0057] Specifically, the key points of the scene include: landing point, water departure point, water entry point and full floating point.

[0058] The landing point is the key point where the track first touches the shore during the landing process of the amphibious unmanned vehicle. After this point, the amphibious unmanned vehicle should simultaneously control the land actuator and the water actuator to achieve the landing action.

[0059] The departure point is the key point during the landing process of the amphibious unmanned vehicle where the water actuator completely fails (the water gate completely leaves the water surface). After this point, the amphibious unmanned vehicle should only control the land actuator to complete the landing mission.

[0060] The entry point is the key point when the water actuator starts to work (the water gate starts to contact the water surface) during the launching process of the amphibious unmanned vehicle. After this point, the amphibious unmanned vehicle should simultaneously control the land actuator and the water actuator to realize the entry action.

[0061] The full floating point is the key point when the tracks are completely submerged in water during the launching process of the amphibious unmanned vehicle. After this point, the amphibious unmanned vehicle should only control the water actuators to complete the water navigation mission.

[0062] Compared with the existing technology, the collaborative control method of the water and land actuators of the amphibious unmanned vehicle provided in this embodiment obtains the suspension state data and attitude angle data of the amphibious unmanned vehicle at each moment; predicts the water model mismatch probability at each current moment, determines whether the current moment is a scene key point based on the predicted water model mismatch probability, and confirms the solution mode according to whether the current moment is a scene key point and the predicted water model mismatch probability to solve the actuator action instructions of the amphibious unmanned vehicle, thereby realizing collaborative stable process control of the water actuator and the land actuator.

[0063] Taking into account actuator response delays and control process smoothness, the water model mismatch probability is predicted to avoid the lag problem of critical point judgment caused by using the current model mismatch probability. For example, during landing, it is necessary to shift gears and accelerate some time in advance. If the current water model mismatch probability is used to determine that the landing point is the landing point, the vehicle will be stuck at the touchdown point.

[0064] The suspension state and attitude angle changes near key points of an amphibious unmanned vehicle are also important indicators for the vehicle's perception of the characteristics of water-land transition scenarios. For example, during water entry, the tracks gradually leave the ground, causing changes in suspension force and visible signatures on the suspension displacement sensor. During landing, as the tracks gradually touch the ground, the influence of water flow on the attitude angle gradually weakens. Therefore, the suspension state data and attitude angle data are extracted as features.

[0065] Specifically, the recurrent neural network is trained in the following way:

[0066] The speed, engine speed, suspension state data, and attitude angle data of the amphibious unmanned vehicle are obtained at each moment in the entire history. The water model mismatch probability at each moment is calculated based on the speed and engine speed. The features of each moment are calculated based on the suspension state data and attitude angle data at each moment. The training sample set is constructed by taking the features of the k+1 moment as the input data of a sample and the average water model mismatch probability of the k+1 moments after the k+1 moment as the sample label.

[0067] Constructing a recurrent neural network model; training the constructed recurrent neural network model based on the training sample set to obtain a trained recurrent neural network model.

[0068] During implementation, the entire historical process includes the entire process of the unmanned vehicle moving from land to water and from water to road.

[0069] During implementation, the water model mismatch probability at each moment is calculated based on the speed and engine speed in the following way:

[0070]

[0071] Where v represents speed, n represents engine speed, and λ w represents the kinematic model coefficient of the water working condition, λ l Represents the kinematic model coefficient of onshore conditions. l and λ w is the known quantity calibrated in the experiment, corresponding to the conversion coefficients of vehicle speed on land and water and engine speed, and λ l >λ w .

[0072] Specifically, the features at each moment are calculated based on the suspension state data and attitude angle data at each moment, including:

[0073] For each moment, the time period from the sth moment before the moment to the moment is the characteristic time period corresponding to the moment;

[0074] The maximum value of the suspension state data, the average value of the suspension state data, the variance of the suspension state data, the maximum value of the attitude angle data, the average value of the attitude angle data and the variance of the attitude angle data in the characteristic time period corresponding to the moment are calculated as the features corresponding to the moment.

[0075] For example, for time t, the time period between time ts and time t is the feature time period corresponding to that time. The maximum value, average value, variance of the suspension state data, maximum value, average value, and variance of the attitude angle data within the feature time period are used as the features corresponding to time t, and the features corresponding to each time are obtained.

[0076] During implementation, the time step length of the sample is k+1, that is, the features of k+1 moments are used as the input data of one sample, and the trained recurrent neural network is used to predict the future water model adaptation probability. Therefore, the average water model mismatch probability of k+1 moments after the k+1 moments is used as the label to obtain a sample.

[0077] At each current moment of the water-land working condition, the water model mismatch probability is predicted based on the trained recurrent neural network according to the suspension state data and attitude angle data at the current moment and historical moments.

[0078] Specifically, the historical moments are the k moments before the current moment. For example, if the current moment is t, then the moments tk, t-(k-1), ..., t-1 are taken as the historical moments.

[0079] The features corresponding to the current moment and each historical moment are calculated according to the above-mentioned feature calculation method to form a feature sequence. The feature sequence is input into the trained recurrent neural network to predict the water model mismatch probability, that is, the average water model mismatch probability in the future k+1 moments.

[0080] After the water model mismatch probability is predicted, determine whether the current point is a scene key point based on the predicted water model mismatch probability:

[0081] If the water model mismatch probability predicted at the current moment is greater than 0, and the water model mismatch probability predicted at the previous moment is equal to 0, then the current moment is the landing point;

[0082] If the mismatch probability of the water model predicted at the current moment is greater than 90%, and the mismatch probability of the water model predicted at the previous moment is less than or equal to 90%, then the current point is the water departure point;

[0083] If the mismatch probability of the water model predicted at the current moment is less than 100%, and the mismatch probability of the water model predicted at the previous moment is equal to 100%, then the current moment is the entry point;

[0084] If the mismatch probability of the water model predicted at the current moment is less than 10%, and the mismatch probability of the water model predicted at the previous moment is greater than or equal to 10%, the current state is full floating point.

[0085] The actuators include gears, throttle, water direction control mechanism and land direction control mechanism. The water direction control mechanism includes a steering wheel, and the land direction control mechanism includes a steering rudder and a water gate.

[0086] The actuator action instructions of the amphibious unmanned vehicle are calculated based on whether the current scene is a key point and the predicted water model mismatch probability, including:

[0087] S41, determining a current solution mode and a gear instruction according to whether the current scene is a key point, wherein the solution mode includes a water mode, a water-land mode, and a land mode;

[0088] S41, calculating the throttle opening according to the current solution mode and the predicted water model mismatch probability;

[0089] S43. If the current mode is land mode, the control instructions of the land direction control mechanism are determined according to the expected curvature; if the current mode is water mode, the control instructions of the water direction control mechanism are determined according to the expected curvature; if the current mode is land-water mode, the control instructions of the land direction control mechanism and the water direction control mechanism are determined according to the expected curvature.

[0090] During implementation, the solution modes include overwater mode, land-water mode, and land-based mode. During the launch process, the land-based mode is used before the water entry point, and only the motion commands of the land-based actuators are controlled. The land-water mode is used between the water entry point and the full floating point, and the motion commands of the land-based actuators and the overwater actuators are calculated simultaneously. After the full floating point, the overwater mode is used, and only the motion commands of the overwater actuators are controlled.

[0091] Similarly, during the landing process, the mode is on water before the landing point, and only the motion instructions of the on-water actuator need to be controlled; the mode is on land between the landing point and the full floating point, and the motion instructions of the on-land actuator and the motion instructions of the on-water actuator need to be calculated at the same time; after leaving the water point, the mode is on land, and only the motion instructions of the on-land actuator need to be controlled.

[0092] Specifically, the current solution mode and gear position instruction are determined based on whether the current scene is a key point, including:

[0093] If the current point is the water entry point, the solution mode is the water-land mode;

[0094] If the current mode is full floating point, the solution mode is water mode, and the gear command is the gear shift command;

[0095] If the current point is a landing point, the solution mode is the land-water mode, and the gear command is the gear engagement command;

[0096] If the current point is out of water, the solution mode is on land mode.

[0097] If the current point is a water entry point or a landing point, the solution mode is set to the water-land mode; if the current point is a full floating point, the mode is set to the water mode; if the current point is a water departure point, the mode is set to the land mode.

[0098] Considering that the rotation of the tracks on water will consume some power, resulting in insufficient propeller power, the gear should be disengaged at the full floating point; considering that the driving resistance increases after the tracks touch the ground and the propeller is not sufficient to push the vehicle ashore, the gear should be engaged at the landing point.

[0099] Specifically, the throttle opening is calculated based on the current solution mode and the predicted water model mismatch probability, including:

[0100] Determine the conversion coefficient between engine speed and unmanned vehicle speed based on the predicted water model mismatch probability;

[0101] If the current solution mode is the land mode or the water-land mode, the throttle opening is calculated according to the expected speed and the conversion coefficient;

[0102] If the current solution mode is the water mode, the expected speed is corrected, and the throttle opening is calculated based on the corrected expected speed and the conversion coefficient.

[0103] Specifically, the conversion coefficient between the engine speed and the unmanned vehicle speed is determined using the following formula based on the predicted water model mismatch probability:

[0104]

[0105] Where λ represents the conversion coefficient between engine speed and unmanned vehicle speed, represents the predicted water model mismatch probability, λ l represents the longitudinal kinematic model coefficient for onshore conditions, λ w is the longitudinal kinematic model coefficient for water working conditions.

[0106] If the current mode is land mode or water mode, the throttle opening is directly calculated according to the expected speed and the conversion coefficient.

[0107] Specifically, the throttle opening is calculated using the following formula:

[0108]

[0109] Δn=n des -n

[0110] Θ(k+1)=Θ(k)+pΔn

[0111] Among them, v des represents the expected speed, n des represents the desired engine speed, n represents the current engine speed, Θ(k) represents the current throttle opening, Θ(k+1) represents the calculated throttle opening, and p represents the coefficient.

[0112] The desired speed is determined by the amphibious unmanned vehicle's planning. After the throttle opening is calculated, the corresponding control command is generated to control the throttle opening.

[0113] A water speedometer, based on the Bernoulli principle, can accurately measure vehicle speed in still water. However, at the same vehicle speed in upstream conditions, the relative water velocity is higher and the pressure is lower than in still water, resulting in the speed data collected by the water speedometer being lower than the actual vehicle speed. Similarly, in downstream conditions, the water speed data collected by the water speedometer is higher than the actual vehicle speed.

[0114] Therefore, if the current solution mode is the water mode, the expected speed needs to be corrected first, and the throttle opening is calculated based on the corrected expected speed and the conversion coefficient.

[0115] Specifically, the expected speed is corrected, including:

[0116] Calculate the theoretical flow velocity characteristics and calculated flow velocity characteristics within the characteristic time period corresponding to the current moment;

[0117] Obtaining a correction coefficient based on a trained correction coefficient prediction model according to the theoretical flow velocity characteristics and the calculated flow velocity characteristics;

[0118] Correct the expected speed according to the correction factor.

[0119] Specifically, the trained neural network model is obtained in the following way:

[0120] Collect the actual speed and water speed meter speed of the amphibious unmanned vehicle at each moment in the entire historical process to calculate the theoretical flow rate; calculate the flow rate based on the expected speed and actual speed at each moment;

[0121] For each moment, the time period from the sth moment before the moment to the moment is the characteristic time period corresponding to the moment; the theoretical flow velocity maximum value, average value and variance of the characteristic time period corresponding to the moment are calculated as the theoretical flow velocity feature corresponding to the moment; the flow velocity maximum value, average value and attitude variance are calculated as the calculated flow velocity feature corresponding to the moment;

[0122] The theoretical flow velocity characteristics and calculated flow velocity characteristics at each moment are used as the input of a sample, and the ratio of the expected velocity at the moment to the time velocity is used as the label to construct a training sample set;

[0123] A neural network model is constructed, and the neural network model is trained based on the constructed training sample set to obtain a trained correction coefficient prediction model.

[0124] During implementation, the vehicle speed collected by the inertial navigation is taken as the real vehicle speed v IMU , the vehicle speed v collected by the water speed meter WS , then the physical quantity that can quantitatively reflect the water flow velocity characteristics is the theoretical flow velocity v t for:

[0125] v t =v IMU -vWS

[0126] Assuming that the speed control deviation is mainly caused by the water flow velocity, the physical quantity that can reflect the deviation between the desired speed and the actual speed of the vehicle is the calculated flow velocity v c Expressed as:

[0127] v c =v des -v IMU

[0128] During implementation, a multi-layer perceptron model may be used to construct a neural network model.

[0129] According to the above method, the theoretical flow velocity characteristics and calculated flow velocity characteristics in the characteristic time period corresponding to the current moment are calculated, and the correction coefficient is obtained by inputting the trained correction coefficient prediction model.

[0130] Correct the expected speed v′ according to the correction coefficient des =γ*v des ,γ represents the correction coefficient, v des represents the expected velocity, v′ des = represents the corrected desired speed. Substitute the corrected desired speed into the above formula for calculating the throttle opening to calculate the throttle opening. Generate the corresponding control command to control the throttle opening.

[0131] Control the direction control mechanism according to the current solution mode.

[0132] Specifically, if the current mode is land mode, the control instructions of the land direction control mechanism are determined according to the expected curvature; if the current mode is water mode, the control instructions of the water direction control mechanism are determined according to the expected curvature; if the current mode is land-water mode, the control instructions of the land direction control mechanism and the water direction control mechanism are determined according to the expected curvature.

[0133] Although each steering actuator and its control process are relatively independent, over-execution may occur when multiple actuators work together to independently achieve the same steering target. However, considering the different characteristics of the land and water scenarios in which amphibious unmanned vehicles operate (such as track contact area, steering rudder and sluice gate immersion depth), the responsiveness of the two sets of steering actuators on land and water will also vary (it is possible that a single actuator cannot achieve the desired curvature, but multiple actuators working together can achieve the desired effect). Therefore, no correction is made to the desired curvature of the actuator. In addition, land and water conditions usually perform straight-line driving tasks such as landing or launching, which require lower steering control accuracy. Therefore, only a brief explanation of the solution of the steering control variable is provided.

[0134] If the current mode is land mode, the land direction control mechanism - the steering wheel angle = steering wheel steering coefficient * desired curvature, generates corresponding control instructions to achieve the steering action. The steering coefficient is calibrated through experiments.

[0135] If the current mode is overwater, the steering rudder angle = steering rudder steering coefficient * desired curvature. This generates corresponding control commands to achieve steering action. If the desired curvature exceeds the steering rudder's turning threshold, the desired curvature determines the steering direction and controls the corresponding water gate to close. The steering coefficient is calibrated through experiments.

[0136] When the desired curvature exceeds the steering threshold of the steering rudder, the water gate is used to compensate for the lack of responsiveness of the steering rudder. For example, the turning direction is determined by the positive or negative value of the desired curvature, and the corresponding water gate is closed. For example, if it is determined to be a left turn, the left water gate is closed.

[0137] If the current mode is land-water mode, the steering wheel and rudder are solved at the same time to determine the corresponding control instructions, so as to avoid the cross-border working conditions affecting the response state of the actuator.

[0138] Those skilled in the art will appreciate that all or part of the process steps of the above-described embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, such as a magnetic disk, an optical disk, a read-only memory, or a random access memory.

[0139] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for collaborative control of amphibious actuators of an amphibious unmanned vehicle, characterized in that: The following steps are involved: Obtain the suspension state data and attitude angle data of the amphibious unmanned vehicle at each moment; At the current moment, the water model mismatch probability is predicted based on the trained recurrent neural network according to the suspension state data and attitude angle data at the current moment and historical moments; the water model mismatch probability is used to characterize the response state of the actuator of the amphibious unmanned vehicle in water and land conditions; Determine whether the current point is a key point in the scene based on the predicted water model mismatch probability; The key points of the scene include: landing point, water departure point, water entry point and full floating point; Calculate the actuator action instructions of the amphibious unmanned vehicle based on whether the current scene is a key point and the predicted water model mismatch probability; The actuators of the amphibious unmanned vehicle include gears, throttle, water direction control mechanism and land direction control mechanism; The actuator action instructions of the amphibious unmanned vehicle are calculated based on whether the current scene is a key point and the predicted water model mismatch probability, including: Determine the current solution mode and gear instruction according to whether the current scene is a key point, wherein the solution mode includes water mode, land-water mode and land mode; Calculate the throttle opening according to the current solution mode and the predicted water model mismatch probability; If the current mode is land, the control instructions of the land direction control mechanism are determined according to the expected curvature; if the current mode is water, the control instructions of the water direction control mechanism are determined according to the expected curvature; if the current mode is land-water mode, the control instructions of the land direction control mechanism and the water direction control mechanism are determined according to the expected curvature; Calculate the throttle opening based on the current solution mode and the predicted water model mismatch probability, including: Determine the conversion coefficient between engine speed and unmanned vehicle speed based on the predicted water model mismatch probability; If the current solution mode is the land mode or the water-land mode, the throttle opening is calculated according to the expected speed and the conversion coefficient; If the current solution mode is the water mode, the desired speed is corrected, and the throttle opening is calculated based on the corrected desired speed and the conversion coefficient; Based on the predicted water model mismatch probability, the conversion coefficient between the engine speed and the unmanned vehicle speed is determined using the following formula: Where λ represents the conversion coefficient between engine speed and unmanned vehicle speed, represents the predicted water model mismatch probability, λ l represents the kinematic model coefficient of the onshore working condition, λ w is the kinematic model coefficient for the water working condition; the kinematic model coefficient for the land working condition corresponds to the conversion coefficient between the vehicle speed and the engine speed under the land working condition, and the kinematic model coefficient for the water working condition corresponds to the conversion coefficient between the vehicle speed and the engine speed under the water working condition; The throttle opening is calculated using the following formula: Δn=n des -n Θ(k+1)=Θ(k)+pΔn Among them, v des represents the expected speed, n des represents the desired engine speed, n represents the current engine speed, Θ(k) represents the current throttle opening, Θ(k+1) represents the calculated throttle opening, and p represents the coefficient; The recurrent neural network is trained in the following way: The speed, engine speed, suspension state data, and attitude angle data of the amphibious unmanned vehicle are obtained at each moment in the entire history. The water model mismatch probability at each moment is calculated based on the speed and engine speed. The features of each moment are calculated based on the suspension state data and attitude angle data at each moment. The training sample set is constructed by taking the features of the k+1 moment as the input data of a sample and the average water model mismatch probability of the k+1 moments after the k+1 moment as the sample label. Constructing a recurrent neural network model; training the constructed recurrent neural network model based on the training sample set to obtain a trained recurrent neural network model; The features of each moment are calculated based on the suspension state data and attitude angle data at each moment, including: For each moment, the time period from the sth moment before the moment to the moment is the characteristic time period corresponding to the moment; Calculate the maximum value of the suspension state data, the average value of the suspension state data, the variance of the suspension state data, the maximum value of the attitude angle data, the average value of the attitude angle data, and the variance of the attitude angle data in the characteristic time period corresponding to the moment as the features corresponding to the moment; The water model mismatch probability at each moment is calculated based on the speed and engine speed in the following way: Where v represents speed, n represents engine speed, and λ w represents the kinematic model coefficient of the water working condition, λ l Represents the kinematic model coefficient of onshore conditions.

2. The method for cooperative control of amphibious actuators of an amphibious unmanned vehicle according to claim 1, characterized in that: Determine whether the current point is a key point in the scene based on the predicted water model mismatch probability, including: If the water model mismatch probability predicted at the current moment is greater than 0, and the water model mismatch probability predicted at the previous moment is equal to 0, then the current moment is the landing point; If the mismatch probability of the water model predicted at the current moment is greater than 90%, and the mismatch probability of the water model predicted at the previous moment is less than or equal to 90%, then the current point is the water departure point; If the mismatch probability of the water model predicted at the current moment is less than 100%, and the mismatch probability of the water model predicted at the previous moment is equal to 100%, then the current moment is the entry point; If the mismatch probability of the water model predicted at the current moment is less than 10%, and the mismatch probability of the water model predicted at the previous moment is greater than or equal to 10%, the current state is full floating point.

3. The method for coordinated control of amphibious actuators of an amphibious unmanned vehicle according to claim 1, characterized in that: The current solution mode and gear position instructions are determined based on whether the current scene is a key point, including: If the current point is the water entry point, the solution mode is the water-land mode; If the current mode is full floating point, the solution mode is water mode, and the gear command is the gear shift command; If the current point is a landing point, the solution mode is the land-water mode, and the gear command is the gear engagement command; If the current point is out of water, the solution mode is on land mode.

4. The method for coordinated control of amphibious actuators of an amphibious unmanned vehicle according to claim 1, characterized in that: Corrections to the desired speed include: Calculate the theoretical flow velocity characteristics and calculated flow velocity characteristics within the characteristic time period corresponding to the current moment; Obtaining a correction coefficient based on a trained correction coefficient prediction model according to the theoretical flow velocity characteristics and the calculated flow velocity characteristics; Correct the expected speed according to the correction factor.

Images