Amphibious vehicle working condition autonomous switching decision-making method and device

Abstract

The invention discloses an amphibious vehicle working condition autonomous switching decision-making method and device. The method comprises the following steps: constructing a multi-dimensional sensing vector corresponding to each time point according to sensing data collected by each sensor of the amphibious vehicle in real time; according to the multi-dimensional perception vector of the time point, adopting a pre-trained switching probability prediction model to predict respective execution probabilities of a plurality of switching instructions corresponding to the time point; determining candidate switching instructions of the time points according to the execution probabilities corresponding to the time points; and performing consistency verification on the candidate switching instruction of each time point in a preset sliding time window, and determining a target switching instruction of a target time point. According to the invention, the accuracy and robustness of working condition switching of the amphibious vehicle in the complex dynamic change amphibious environment can be improved.

Description

Technical Field

[0001] This invention relates to the field of amphibious vehicle technology, and in particular to a method and device for autonomous switching decision-making of amphibious vehicle operating conditions. Background Technology

[0002] Amphibious vehicles are a type of multi-functional mobile equipment capable of operating flexibly in both terrestrial and water environments, and are widely used in both military and civilian fields such as rescue, reconnaissance, and combat. To ensure the successful execution of missions, amphibious vehicles must be able to switch quickly and accurately between land and water to adapt to the varying demands of different terrains on vehicle operation. Existing autonomous switching methods are based on sensor data or vehicle location. These methods collect data such as water depth, terrain slope, and road conditions, and combine this with preset rules to determine the timing of the switch, thereby improving the efficiency and accuracy of land-water transitions to a certain extent.

[0003] While existing methods have made some progress in solving water-to-land switching problems in some simple scenarios, they still have significant limitations in complex environments. First, switching methods based on single-time-series data are susceptible to noise or outliers, lacking a holistic understanding of the dynamic evolution of the environment and vehicle status, which can easily lead to misjudgments and inaccurate or untimely switching. Second, the pre-set rules are usually fixed and difficult to adapt to changing environments and complex task requirements. For example, in environments with large waves or significant underwater topographic changes, fixed rules may fail to make effective switching decisions, thus affecting vehicle safety and efficiency. Furthermore, existing methods model water-to-land switching decisions as a classification problem, relying on single-time-slice data for judgment, lacking a comprehensive understanding of continuous environmental states, and failing to provide stable switching decisions. Especially in changing and complex environments, they are susceptible to interference from instantaneous noise or errors, leading to misjudgments. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a method and apparatus for autonomous switching of operating conditions for amphibious vehicles, aiming to improve the accuracy and robustness of amphibious vehicles switching operating conditions in complex and dynamically changing amphibious environments, and to provide more intelligent and reliable support for the autonomous switching of amphibious vehicles.

[0005] To achieve the above objectives, according to a first aspect of the present invention, an autonomous switching decision-making method for amphibious vehicles is provided, comprising:

[0006] Based on the environmental perception data of amphibious vehicles, construct multi-dimensional perception vectors corresponding to each time point;

[0007] Based on the multidimensional sensing vector at the time point, and using a pre-trained handover probability prediction model, the execution probability of each of the multiple handover instructions corresponding to the time point is predicted.

[0008] The candidate switching instruction for the time point is determined based on the execution probabilities corresponding to the time point;

[0009] The consistency of the candidate switching instructions at each time point in the preset sliding time window is checked to determine the target switching instruction at the target time point. The target switching instruction is executed by the amphibious vehicle at the target time point.

[0010] Furthermore, the environmental perception data includes at least two items: water depth data, track vertical load, coastline distance, and suspension deformation; the switching probability model is a GMM-HMM switching probability model, and each switching probability model corresponds to one of the switching commands of the amphibious vehicle, the switching commands including switching to water, switching to land, switching to amphibious transition, and not switching.

[0011] Furthermore, based on the multidimensional sensing vector at the stated time point, and using a pre-trained handover probability prediction model, the execution probability of each of the multiple handover instructions corresponding to the stated time point is predicted, including:

[0012] The execution probability P(o) corresponding to the s-th switching instruction is calculated according to the following formula. T |λ s ):

[0013]

[0014] Among them, o t and z t These are the multidimensional sensing vector and the switching instruction at the t-th time point, respectively, t = 1, 2, ..., T; λ s These are the model parameters obtained after pre-training the s-th GMM-HMM model. It is the probability distribution of the initial handover command z1 obtained from the s-th GMM-HMM handover probability model. The switching command z is obtained from the s-th GMM-HMM switching probability model. t-1 and the switching command z t The transition probability between them The switching command z is obtained from the s-th GMM-HMM switching probability model. t The following describes the multidimensional sensing vector o t The probability of observation.

[0015] Furthermore, determining the candidate switching instruction for each time point based on the execution probabilities corresponding to that time point includes:

[0016] The candidate switching instruction z at the t-th time point is calculated according to the following formula. t :

[0017]

[0018] Furthermore, a consistency check is performed on the candidate switching instructions at each time point within the preset sliding time window to determine the target switching instruction at the target time point, including:

[0019] The sliding time window T is statistically analyzed. w The probability of occurrence P(z) of each of the candidate switching instructions within the specified range t =z), z∈{1,2,3,4}, t∈T w ;

[0020] Candidate switching instructions with a probability greater than the consistency threshold are selected based on the probability of occurrence of each of the aforementioned events.

[0021] In response to selecting one or more candidate switching instructions that are greater than the consistency threshold, the candidate switching instruction corresponding to the highest occurrence probability is executed.

[0022] Furthermore, the pre-training of the GMM-HMM switching probability model includes:

[0023] In various water-land switching scenarios, historical data sequences composed of the multi-dimensional sensing vectors are collected;

[0024] For each of the aforementioned operating conditions, the historical data sequence is labeled with an instruction tag corresponding to the switching instruction. The historical data sequence is then segmented according to the instruction tag to obtain the training dataset corresponding to each of the aforementioned switching instructions.

[0025] Based on the training dataset corresponding to the switching instruction, the model parameters of the GMM-HMM model to be trained are initialized, and the GMM-HMM model is iteratively optimized using the EM algorithm to obtain the GMM-HMM switching probability model.

[0026] According to a second aspect of the present invention, an autonomous switching decision-making device for amphibious vehicles is provided, comprising:

[0027] The module is used to construct multi-dimensional perception vectors corresponding to each time point based on the environmental perception data of amphibious vehicles.

[0028] The prediction module is used to predict the execution probability of each of the multiple switching instructions corresponding to the time point based on the multidimensional sensing vector at the time point and by using a pre-trained switching probability prediction model.

[0029] The filtering module is used to filter out candidate switching instructions for the time point according to the execution probabilities corresponding to the time point;

[0030] The decision module is used to perform consistency verification on the candidate switching instructions at each time point in the preset sliding time window, and determine the target switching instruction at the target time point. The target switching instruction is executed by the switching control module of the amphibious vehicle at the target time point.

[0031] According to a third aspect of the present invention, a computer-readable storage medium is provided having a program stored thereon that, when executed by a processor, implements the steps of the amphibious vehicle autonomous switching decision method as described in the first aspect of the present invention.

[0032] According to a fourth aspect of the present invention, a terminal device is provided, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the amphibious vehicle autonomous switching decision method as described in the first aspect of the present invention.

[0033] According to a fifth aspect of the present invention, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the steps in the amphibious vehicle autonomous switching decision method as described in the first aspect of the present invention.

[0034] The embodiments of the present invention have at least one of the following advantages or beneficial effects:

[0035] This invention improves the accuracy and robustness of amphibious vehicles switching operating conditions in complex and dynamically changing water and land environments by constructing a multi-dimensional perception system and using a pre-trained switching probability prediction model to predict the execution probability of candidate switching commands at various time points, thereby obtaining candidate switching commands and performing consistency verification through a sliding time window.

[0036] This invention employs a Gaussian Mixture Hidden Markov Model (GMM-HMM) to model the relationship between each switching command and the environmental state of an amphibious vehicle. By leveraging the observational modeling capabilities of the Gaussian Mixture Model (GMM) and the state transition characteristics of the Hidden Markov Model (HMM), it is possible to perform deep learning and analysis on the dynamic evolution of each switching command and the environmental state. Compared to traditional classification methods based on single-time-slice data, this invention can capture the continuous changing trend of the environmental state, thereby significantly improving the accuracy of switching command decisions.

[0037] The sliding time window mechanism of this invention caches and accumulates prediction results over a short period of time, and triggers a switching command only when the prediction results at multiple time points meet the consistency threshold. This mechanism effectively avoids misjudgments caused by instantaneous noise or data jitter, and significantly improves the robustness of decision-making, especially in complex environments such as areas with large waves or significant changes in underwater topography, where sensor data may be affected by noise or outliers.

[0038] The various technical solutions described above in this invention can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained through the details specifically pointed out in the description and drawings. Attached Figure Description

[0039] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0040] Figure 1 This is a schematic diagram of the main process of an autonomous switching decision-making method for amphibious vehicles according to an embodiment of the present invention.

[0041] Figure 2 This is a schematic diagram illustrating the principle of the GMM-HMM switching probability model in this embodiment of the invention.

[0042] Figure 3 This is a schematic diagram of the main process of pre-training the GMM-HMM switching probability model in an embodiment of the present invention.

[0043] Figure 4 This is a schematic diagram illustrating the principle of consistency verification based on a sliding time window in an embodiment of the present invention.

[0044] Figure 5 This is a schematic diagram of the main components of an amphibious vehicle autonomous switching decision-making device according to an embodiment of the present invention.

[0045] Figure 6 This is a schematic diagram of the composition of a terminal device according to an embodiment of the present invention. Detailed Implementation

[0046] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0047] Example 1

[0048] Figure 1This is a schematic diagram of the main flow of a method for autonomous switching of operating conditions for amphibious vehicles according to an embodiment of the present invention. Figure 1 As shown, the autonomous switching decision method for amphibious vehicle operating conditions in this embodiment of the present invention includes the following steps S101 to S104.

[0049] Step S101: Construct multi-dimensional perception vectors corresponding to each time point based on the environmental perception data of the amphibious vehicle.

[0050] Step S102: Based on the multidimensional sensing vector at the time point, and using a pre-trained switching probability prediction model, predict the execution probability of each of the multiple switching instructions corresponding to the time point.

[0051] Step S103: Determine the candidate switching instruction for the time point based on the execution probabilities corresponding to the time point;

[0052] Step S104: Perform consistency verification on the candidate switching instructions at each time point in the preset sliding time window, and determine the target switching instruction at the target time point. The target switching instruction is executed by the amphibious vehicle at the target time point.

[0053] Understandably, in this embodiment and some embodiments of the present invention, the environmental perception data is obtained through continuous observation and analysis of the environmental state of the amphibious vehicle by various sensors. The environmental perception data includes water depth data, track vertical load, coastline distance, and suspension deformation, etc.

[0054] Specifically, in this embodiment and some embodiments of the present invention, to distinguish the different operating conditions of amphibious vehicles, data collection requires the support of multiple sensors, including at least two, to accurately capture the interaction between the vehicle and the environment. The following are some of the sensors used in some embodiments of the present invention and their environmental perception data:

[0055] 1) Depth gauge. Provides water depth information at the vehicle's location, denoted by the symbol 'd', and measured in meters. This environmental perception data helps identify whether the vehicle is in water or in a water-land interface area. By observing changes in water depth, it's possible to infer the vehicle's tendency to gradually enter or leave the water area.

[0056] 2) Track Vertical Load Sensor. Provides information on the track contact force, denoted by the symbol F, and measured in Newtons. This environmental sensing data helps monitor the force on the track under different ground contact conditions. In the transition zone where the vehicle gradually leaves or contacts the ground, the track vertical load can directly reflect the actual force between the track and the contact surface, thus determining whether the vehicle has entered water or returned to land.

[0057] 3) GPS and Inertial Navigation System (INS). Used to estimate the distance between the vehicle and the coastline, providing reference information for relative position. The distance between the current position and the coastline is represented by the symbol 's', and the unit is meters. The change in the distance to the coastline as the vehicle approaches or moves away from the land-water interface can serve as a visual indicator, helping the system determine the vehicle's location at the land-water interface.

[0058] 4) Suspension Deformation Sensor. This sensor monitors the suspension deformation of the vehicle under different terrains and driving conditions. This variable indirectly reflects the vertical load the vehicle experiences during driving, denoted by the symbol x, and measured in millimeters. When the vehicle is traveling on land, the suspension deformation is larger due to the full effect of gravity; while in water, buoyancy alleviates some of the load, significantly reducing the suspension deformation. Therefore, changes in suspension deformation can help determine whether the vehicle is in contact with the ground, thus identifying the current driving environment. This also provides crucial reference for mode switching in areas where land and water meet.

[0059] Understandably, the pre-trained handover probability prediction model learns the sequential features of the collected environmental data, models the relationship between handover commands and environmental states as latent variables, and identifies suitable handover times. Furthermore, to further improve the system's stability under drastic environmental fluctuations, this embodiment of the invention introduces a sliding time window mechanism in the decision-making process to mitigate the impact of single-time-point jitter on handover judgment, thereby making handover decisions more accurate.

[0060] Furthermore, in this embodiment and some embodiments of the present invention, the switching probability model is a GMM-HMM switching probability model, which is trained based on a GMM-HMM (Gaussian Mixture Hidden Markov Model). Each switching probability model corresponds to one of the switching commands of the amphibious vehicle, including switching to water, switching to land, switching to a water-land transition, and not switching.

[0061] 1) Switch to Water: When the vehicle is detected to be in a fully buoyant state, meaning all tracks are off the ground and completely floating, the system should switch to water driving mode to adapt to the needs of the aquatic environment. For ease of discussion later, this will be referred to as switch command 1.

[0062] 2) Switch to Land: When the vehicle has completely returned to land, all tracks (or wheels) are in contact with the ground, and the vehicle body is out of the water, it should switch to land driving mode to adapt to the needs of the ground environment. This is called switch command 2.

[0063] 3) Switch to Water-Land Transition: When the vehicle is in the water-land transition zone, part of the tracks are in contact with the ground, and part of the vehicle body is floating in the water. In this state, the system should switch to water-land transition mode to adapt to the special working conditions of the transition zone, which is recorded as switch command 3.

[0064] 4) No Switch: The current vehicle status is stable and no mode switching is required. This status helps the system avoid unnecessary switching due to environmental noise or sensor errors, and is denoted as Switching Command 4.

[0065] Specifically, in this embodiment and some embodiments of the present invention, the water depth d of the amphibious vehicle at the current time t is collected in real time. t Track vertical load sensor reading F t The distance s between the current location and the coastline t and the reading x of the suspension deformation sensor t Construct the multidimensional perception vector at the current time t, i.e., o t ={d t ,F t ,s t ,x t}, input the four trained GMM-HMM switching probability models.

[0066] Understandably, in this embodiment and some embodiments of the present invention, the water-land switching decision is divided into four types (switching to water, switching to land, switching to a water-land transition, and no switching), and an independent GMM-HMM switching probability model is trained for each type to learn the relationship between the sequence features of the collected environmental data and the switching command, which can adapt to different task requirements and environmental conditions. For example, in the water-land interface area, the system can choose to enter the water-land transition mode according to actual needs, thereby avoiding the risks that may be caused by directly switching from the land mode to the water mode.

[0067] Figure 2 This is a schematic diagram illustrating the principle of the GMM-HMM switching probability model in this embodiment of the invention. Specifically, as shown... Figure 2 As shown, in this embodiment and some embodiments of the present invention, z1 represents the switching command (or state) at time t=1, o1 represents the multi-dimensional sensing vector at time t=1, o1={d1,F1,s1,x1}, including the water depth data d1, track vertical load F, coastline distance s1, and suspension deformation sensor reading x1 at time t=1. t This indicates the switching instruction at time point t, o t The multidimensional sensing vector represents time point t, where t = 1, 2, 3, ... The switching command z... t The range of values ​​for z is set to a discrete space with four possible values.t ∈{1,2,3,4}, where each number represents a switching instruction. The GMM-HMM model can be defined by the quintuple model parameters λ={π,A,w,μ,Σ}.

[0068] Where, π = {π j} represents the probability distribution of the initial switching instruction, π j π represents the probability that the system is in the state of switching instruction j at the initial time point t=1. j =P(z1=j), which needs to satisfy the normalization condition: Specifically, in this embodiment and some embodiments of the present invention, N=4, which is the total number of switching instructions.

[0069] A is the switching instruction transfer matrix, A = {a ij},a ij =P(z) t =j|z t-1 =i), where, a ij From the switching command z t-1 and the switching command z t The probability of transitions between them must satisfy the normalization condition:

[0070] w is the mixing coefficient vector of the GMM (Gaussian Mixture Model), defined as w = {w...} j,m}. , where w j,m =P(m,z) t =j), w j,m This indicates that under switching instruction j, the mixing weight of the m-th Gaussian distribution component (i.e., the probability of this component being selected) must satisfy the normalization condition: M is the number of Gaussian mixture components corresponding to each switching instruction.

[0071] μ is the mean vector of a Gaussian mixture distribution, defined as follows: μ = {μ j,m}. Where μ j,m This indicates that under switching instruction j, the mean vector of the m-th Gaussian distribution component determines the center position of that Gaussian distribution component.

[0072] Σ is the covariance matrix of a Gaussian mixture distribution, defined as follows: Σ={Σ j,m}. Among them, Σ j,m This represents the covariance matrix of the m-th Gaussian distribution component under switching instruction j, which determines the shape and extent of this component.

[0073] Given the parameters λ = {π, A, w, μ, Σ}, the multidimensional sensing vector o t The conditional probability is modeled by GMM:

[0074]

[0075] in, It is the m-th Gaussian distribution component, defined as:

[0076]

[0077] Specifically, in this embodiment and some embodiments of the present invention, D=4 is the multidimensional sensing vector o. t The dimension. According to the first-order Markov property assumption, the sequence of multidimensional perceptual vectors O = (o1, o2, ..., o...). T The joint probability of ) is:

[0078]

[0079] Where Z = (z1, z2, ..., z T P(z1) is the sequence of switching instructions (latent variable), and P(z2) is the probability distribution of the initial switching instruction z1. t |z t-1 ) is the switching instruction z t-1 and the switching command z t The transition probability between them, P(o t |z t ) is the switching instruction z t The following describes the multidimensional sensing vector o t The probability of observation.

[0080] Specifically, in this embodiment and some embodiments of the present invention, step S102 includes:

[0081] The execution probability P(o) corresponding to the s-th switching instruction is calculated according to the following formula. T |λ s ):

[0082]

[0083] Among them, o t and z t These are the multidimensional sensing vector and the switching instruction at the t-th time point, respectively, t = 1, 2, ..., T; λ s These are the model parameters obtained after pre-training the s-th GMM-HMM model. It is the probability distribution of the initial handover command z1 obtained from the s-th GMM-HMM handover probability model. The switching command z is obtained from the s-th GMM-HMM switching probability model. t-1 and the switching command z t The transition probability between them The switching command z is obtained from the s-th GMM-HMM switching probability model. t The following describes the multidimensional sensing vector o t The probability of observation.

[0084] Where, λ s ={π s A s ,w s ,μ s ,Σ i}, π s A s These are the probability distribution matrix and the switching instruction transition matrix of the initial switching instruction obtained after pre-training the s-th GMM-HMM model, respectively. s μ s , Σ i These are the mixing coefficient vector, mean vector, and covariance matrix of the Gaussian mixture distribution obtained after pre-training the s-th GMM-HMM model;

[0085] Specifically, in this embodiment and some embodiments of the present invention, the pre-training of the GMM-HMM switching probability model includes steps S201 to S203:

[0086] Step S201: Collect historical data sequences composed of the multi-dimensional perception vectors under various water-land switching scenarios;

[0087] Step S202: For the historical data sequence of each working condition, label it with an instruction tag corresponding to the switching instruction; divide the historical data sequence into segments according to the instruction tag to obtain the training dataset corresponding to each switching instruction.

[0088] Step S203: Based on the training dataset corresponding to the switching instruction, initialize the model parameters of the GMM-HMM model to be trained, and use the EM algorithm to iteratively optimize the GMM-HMM model to obtain the GMM-HMM switching probability model.

[0089] Understandably, in this embodiment and some embodiments of the present invention, accurately acquiring and labeling the training data corresponding to different switching commands is crucial for constructing an effective switching probability model. Therefore, the training data needs to cover various land-water switching scenarios, including but not limited to swamps, rivers, lakes, and coastlines, and encompass rich sensor information to achieve comprehensive perception and judgment of vehicle and environmental conditions.

[0090] Specifically, data from various land-water switching scenarios are collected to construct a sample set for training the GMM-HMM model. Data from each scenario is labeled with the corresponding switching commands to distinguish different switching commands, enabling the model to learn the characteristics of different environmental states and to learn effective land-water switching command decisions.

[0091] Understandably, in this embodiment and some embodiments of the present invention, based on the actual switching needs of the vehicle in the water-land transition zone, the corresponding training dataset is divided using instruction labels corresponding to the four switching commands, focusing on whether and when to switch driving modes to support more direct decision-making. To better distinguish the correspondence between environmental states and switching commands, multiple sensors are used during data collection to accurately capture the interaction between the vehicle and the environment.

[0092] Specifically, in this embodiment and some embodiments of the present invention, four GMM-HMM models are trained using the Expectation Maximization (EM) algorithm based on the four training datasets constructed in step S202.

[0093] Specifically, in this embodiment and some embodiments of the present invention, the initial value of the model parameter λ = {π, A, w, μ, Σ} can be set according to the actual working conditions, the range of sensor data, and the physical characteristics of the working conditions.

[0094] Specifically, in this embodiment and some embodiments of the present invention, the E step of the EM algorithm is the expectation step, used to calculate the expected value of the latent variables, including:

[0095] Iterative calculation of the forward probability α of point t being at the switching instruction j t (j), j∈{1,2,3,4}, α t (j)=P(o1,o2,...,o t ,z t =j|λ), where,

[0096]

[0097] Where, π j =P(z1=j) is the probability that the initial switching instruction is switching instruction j, b j (o1) is the probability of observing the first multidimensional sensor vector o1 under the switching instruction j, b j (o t ) is the multi-dimensional sensing vector o under the switching instruction j. t The observation probability distribution, that is, the multidimensional sensing vector o given the switching instruction j. t Probability of occurrence:

[0098]

[0099] Then iteratively calculate the backward probability β of time point t being at the switching instruction j. t (j), β t (j)=P(o t+1 ,o t+2 ,...,o T |z t =j,λ): where,

[0100]

[0101] Then, the posterior probability γ of each switching instruction for each multidimensional sensing vector under the current model parameters λ is calculated. t (j) represents the probability that time point t is under switching instruction j.

[0102]

[0103] Where P(O|λ) is the probability of the sequence O of the multidimensional sensing vector under the parameter λ. Then calculate the multidimensional sensing vector o. t At time t, the probability γ is generated by the m-th Gaussian component and is under switching instruction j. t (j,m).

[0104]

[0105] Calculate the probability of a handover instruction transfer between two points in time:

[0106]

[0107] Where, ξ t (i,j) is the probability that at time t, the user is in switching instruction i and at time t+1, the user is transferred to switching instruction j.

[0108] Specifically, in this embodiment and some embodiments of the present invention, the M step of the EM algorithm is maximization, which involves re-estimating the model parameters λ based on the calculation results in the E step, specifically including:

[0109]

[0110]

[0111] By iterating through the E and M steps until convergence, the trained GMM-HMM switching probability models corresponding to the four switching commands can be obtained.

[0112] Understandably, in this embodiment and some embodiments of the present invention, the present invention uses a Gaussian Mixture Hidden Markov Model (GMM-HMM) to model the relationship between each switching command of the amphibious vehicle and the environmental state separately. By leveraging the observational modeling capability of the Gaussian Mixture Model (GMM) and the state transition characteristics of the Hidden Markov Model (HMM), it is possible to perform deep learning and analysis on the dynamic evolution of each switching command and the environmental state. Compared with traditional classification methods based on single time slice data, the present invention can capture the continuous changing trend of the environmental state, thereby significantly improving the accuracy of switching command decisions.

[0113] Specifically, in this embodiment and some embodiments of the present invention, step S103 includes: calculating the candidate switching instruction z at the t-th time point according to the following formula. t : That is, based on the execution probability of each switching instruction, the switching instruction with the highest execution probability is selected as the candidate switching instruction at each time point.

[0114] Specifically, in this embodiment and some embodiments of the present invention, step S104 includes steps S104a to S104c:

[0115] Step S104a: Calculate the sliding time window T. w The probability of occurrence P(z) of each of the candidate switching instructions within the specified range t =z), z∈{1,2,3,4}, t∈T w ;

[0116] Step S104b: Filter out the candidate switching instructions that are greater than the consistency threshold according to the occurrence probability of each of the above.

[0117] Step S104c: In response to selecting one or more candidate switching instructions that are greater than the consistency threshold, execute the candidate switching instruction corresponding to the highest occurrence probability; otherwise, the switching instruction is not to switch.

[0118] Figure 4 This is a schematic diagram illustrating the principle of consistency verification based on a sliding time window in an embodiment of the present invention. Specifically, as shown... Figure 4 As shown, in this embodiment and some embodiments of the present invention, in step S104a, a fixed-length sliding time window w = 5 is set to cache the candidate switching instructions for the most recent w time points. For each time point, the switching instruction z predicted by GMM-HMM is recorded. t The value range is: switch to water (1), switch to land (2), switch to water-land transition (3), no switch (4). The sliding window stores the most recent T. w Candidate switching instructions at each time point:

[0119] Z t ={z t-w+1 ,z t-w+2 ,...,z t}

[0120] Among them, z t This represents the decision result at time t based on the GMM-HMM handover probability model in step S103. Then, the frequency of occurrence of each candidate handover instruction is statistically analyzed within the sliding time window:

[0121]

[0122] in, It is an indicator function:

[0123]

[0124] Specifically, in this embodiment and some embodiments of the present invention, in step S104b, the consistency threshold η = 80% is set, and in step S104c, the mode switch is only executed when the proportion of a certain decision in the window exceeds the threshold; otherwise, the switching instruction is not to switch (4).

[0125]

[0126] Understandably, in this embodiment and some embodiments of the present invention, in order to improve the stability of the operating condition switching decision, a sliding time window is introduced on the basis of the GMM-HMM switching probability model for switching command prediction, to cache and accumulate short-term decision results, to perform consistency checks on the prediction results at multiple times, and to trigger the switching decision only when the set conditions are met, so as to avoid misjudgment caused by short-term fluctuations and improve the stability and anti-interference ability of the decision.

[0127] The method described in this invention offers higher stability, environmental adaptability, and intelligent decision-making efficiency, demonstrating significant advantages in autonomous water-land switching. First, a Gaussian Mixture Hidden Markov Model (GMM-HMM) is employed to perform deep modeling of the vehicle's environmental state time-series data, enabling continuous tracking and pattern recognition of dynamic environmental changes. This method overcomes the limitations of traditional single-time-slice-based decision-making, identifying potential trends in environmental states and enhancing the system's adaptability in complex scenarios.

[0128] The introduction of a time window mechanism in this embodiment of the invention enables caching and accumulation of instantaneous prediction data. This not only avoids immediate responses to data in each time slice but also aggregates judgment results from multiple moments by setting a time window of a certain length. Through this mechanism, the system can perform consistency checks on the time-series judgment results, triggering a switching command only when the prediction data within the window reaches a certain threshold of consistency, thereby significantly improving the stability and robustness of switching decisions. This mechanism effectively avoids misjudgments caused by short-term fluctuations or abnormal data from sensors, ensuring the accuracy and reliability of autonomous switching decisions in complex environments at land-water boundaries.

[0129] The embodiments of the present invention, through the synergistic application of GMM-HMM and time window mechanism, can realize intelligent switching between amphibious and land modes in uncertain and dynamically changing environments, significantly improving the robustness and environmental adaptability of the system, and meeting the needs of amphibious vehicles for efficient autonomous switching under complex terrain and hydrological conditions.

[0130] Example 2

[0131] Figure 5 This is a schematic diagram of the constituent modules of an amphibious vehicle autonomous switching decision-making device according to an embodiment of the present invention. Figure 5 As shown, the amphibious vehicle autonomous switching decision-making device of this embodiment includes:

[0132] The module is used to construct multi-dimensional perception vectors corresponding to each time point based on the environmental perception data of amphibious vehicles.

[0133] The prediction module is used to predict the execution probability of each of the multiple switching instructions corresponding to the time point based on the multidimensional sensing vector at the time point and by using a pre-trained switching probability prediction model.

[0134] The filtering module is used to filter out candidate switching instructions for the time point according to the execution probabilities corresponding to the time point;

[0135] The decision module is used to perform consistency verification on the candidate switching instructions at each time point in the preset sliding time window, and determine the target switching instruction at the target time point. The target switching instruction is executed by the switching control module of the amphibious vehicle at the target time point.

[0136] Example 3

[0137] like Figure 6As shown, Embodiment 3 of the present invention provides a terminal device, including at least one processor and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps in the amphibious vehicle autonomous switching decision method as described in the first aspect of the present invention.

[0138] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripherals, voltage regulators, and power management circuits, via interfaces, as is well known in the art. Interfaces provide a connection between the bus and the transceiver, such as communication interfaces or user interfaces. The transceiver can be a single component or multiple components, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0139] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0140] Example 4

[0141] Embodiment 4 of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the amphibious vehicle autonomous switching decision method as described in the first aspect of the present invention.

[0142] Those skilled in the art will understand from the foregoing description that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes, but is not limited to, various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic storage devices, and optical storage devices.

[0143] Example 5

[0144] Embodiment 5 of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the amphibious vehicle autonomous switching decision method as described in the first aspect of the present invention.

[0145] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0146] Based on the detailed description of the embodiments of the present invention above, it can be seen that the present invention has the following advantages:

[0147] This invention improves the accuracy and robustness of amphibious vehicles switching operating conditions in complex and dynamically changing water and land environments by constructing a multi-dimensional perception system and using a pre-trained switching probability prediction model to predict the execution probability of candidate switching commands at various time points. The candidate switching commands are then obtained and consistency is verified through a sliding time window to determine the target switching command.

[0148] This invention employs a Gaussian Mixture Hidden Markov Model (GMM-HMM) to model the relationship between each switching command and the environmental state of an amphibious vehicle. By leveraging the observational modeling capabilities of the Gaussian Mixture Model (GMM) and the state transition characteristics of the Hidden Markov Model (HMM), it is possible to perform deep learning and analysis on the dynamic evolution of each switching command and the environmental state. Compared to traditional classification methods based on single-time-slice data, this invention can capture the continuous changing trend of the environmental state, thereby significantly improving the accuracy of switching command decisions.

[0149] The sliding time window mechanism of this invention caches and accumulates prediction results over a short period of time, and triggers a switching command only when the prediction results at multiple time points meet the consistency threshold. This mechanism effectively avoids misjudgments caused by instantaneous noise or data jitter, and significantly improves the robustness of decision-making, especially in complex environments such as areas with large waves or significant changes in underwater topography, where sensor data may be affected by noise or outliers.

[0150] This invention employs a GMM-HMM model to continuously model environmental conditions and combines this with a sliding time window for in-depth analysis, enabling dynamic adaptation to environmental changes in complex water-land interface areas. For example, at water-land interfaces, multi-dimensional sensing data such as vehicle water depth, track vertical load, coastline distance, and suspension deformation frequently change. This invention can capture these changes in real time and generate accurate switching commands, avoiding misjudgments caused by drastic environmental fluctuations.

[0151] This invention uses a sliding time window to accumulate the prediction results of multiple time slices, ensuring that the generation of switching instructions is based on a stable environmental state judgment, thereby avoiding frequent or erroneous switching. This is especially important in areas where vehicles need to switch driving modes with high stability and consistency.

[0152] This invention utilizes a pre-trained GMM-HMM model to perform real-time analysis of the vehicle's environmental state and combines this with a sliding time window mechanism for consistency verification, enabling the generation of more intelligent switching commands. Compared to traditional rule-based decision-making methods, this invention can dynamically adjust decision logic to adapt to changing environmental conditions, thereby significantly improving the intelligence level of switching decisions.

[0153] This invention improves decision-making efficiency by constructing a multi-dimensional perception vector and predicting switching instructions in real time, thus avoiding the complex rule-by-rule matching process. Furthermore, the introduction of a sliding time window further reduces misjudgments caused by data noise, further optimizing the switching process and improving system operating efficiency.

[0154] The above description is only a preferred 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 conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. An amphibious vehicle operating mode autonomous switching decision method, characterized in that, The method comprises the following steps: According to the environmental perception data collected by various sensors of the amphibious vehicle, a multi-dimensional perception vector corresponding to each time point is constructed; According to the multi-dimensional perception vector of the time point and using a pre-trained switching probability prediction model, the execution probability of each of the multiple switching instructions corresponding to the time point is predicted; According to each of the execution probabilities corresponding to the time point, a candidate switching instruction of the time point is determined; The consistency of the candidate switching instructions of each of the time points in a preset sliding time window is verified, and a target switching instruction of a target time point is determined, which is used for the amphibious vehicle to execute at the target time point.

2. The method of claim 1, wherein, The environmental perception data at least includes two of water depth data, track vertical load, coastline distance and suspension deformation variable; the switching probability model is a GMM-HMM switching probability model, each of which corresponds to one of the switching instructions of the amphibious vehicle, and the switching instructions include switching to water, switching to land, switching to water-land transition and no switching.

3. The method of claim 2, wherein, According to the multi-dimensional perception vector of the time point and using a pre-trained switching probability prediction model, the execution probability of each of the multiple switching instructions corresponding to the time point is predicted, which comprises: The execution probability P(o corresponding to the s-th switching instruction is calculated according to the following formula: T |λ s ) : wherein o t and z t are the multi-dimensional perception vector and the switching instruction at the t-th time point, respectively, t = 1, 2,..., T; λ s is the model parameter of the s-th GMM-HMM model obtained through pre-training, is the probability distribution of the initial switching instruction z1 obtained by the s-th GMM-HMM switching probability model, is the transition probability between the switching instruction z t-1 and the switching instruction z t obtained by the s-th GMM-HMM switching probability model, is the observation probability of the multi-dimensional perception vector o t under the switching instruction z t .

4. The method of claim 3, wherein, According to each of the execution probabilities corresponding to the time point, a candidate switching instruction of the time point is determined, which comprises: The candidate switching instruction z at the t-th time point is calculated according to the following formula t :

5. The method of claim 4, wherein, The consistency of the candidate switching instructions of each of the time points in a preset sliding time window is verified, and a target switching instruction of a target time point is determined, which comprises: counting the occurrence probability P(z w of each of the candidate handover instructions within the sliding time window T t = z), z e {1,2,3,4}, t e T w ; According to each of the appearance probabilities, the candidate switching instructions greater than a consistency threshold are screened out; In response to screening out one or more candidate switching instructions greater than the consistency threshold, the candidate switching instruction corresponding to the maximum appearance probability is executed.

6. The method according to any one of claims 2-5, characterized in that, The pre-training of the GMM-HMM switching probability model comprises: Under various water-land switching scenarios, a historical data sequence composed of the multi-dimensional perception vectors is collected; For each of the working conditions, the historical data sequence is labeled with an instruction label corresponding to the switching instruction, and the historical data sequence is segmented according to the instruction label to obtain a training data set corresponding to each of the switching instructions; According to the training data set corresponding to the switching instruction, the model parameters of the GMM-HMM model to be trained are initialized, and the EM algorithm is used to iteratively optimize the GMM-HMM model to obtain the GMM-HMM switching probability model.

7. An amphibious vehicle operating mode autonomous switching decision device, characterized by, The method comprises the following steps: A construction module is configured to construct a multi-dimensional perception vector corresponding to each time point according to environmental perception data of an amphibious vehicle; A prediction module is configured to predict the execution probability of each of the multiple switching instructions corresponding to the time point according to the multi-dimensional perception vector of the time point and using a pre-trained switching probability prediction model; A screening module is configured to screen out a candidate switching instruction of the time point according to each of the execution probabilities corresponding to the time point; A decision module is configured to verify the consistency of the candidate switching instructions of each of the time points in a preset sliding time window, and determine a target switching instruction of a target time point, which is used for a switching control module of the amphibious vehicle to execute at the target time point.

8. A computer-readable storage medium, characterized in that, A computer program product comprising a computer readable medium having stored thereon a computer program which, when executed by a processor, implements the method of any one of claims 1-6.

9. A terminal device, comprising: A computer program product comprising a computer readable medium having stored thereon a computer program which, when executed by a processor, implements the method of any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, A computer program product comprising a computer readable medium having stored thereon a computer program which, when executed by a processor, implements the method of any one of claims 1-6.

Images