Helicopter Short Baseline Navigation Method Based on China's Beidou Pose State Vector

Abstract

The present invention discloses a short baseline navigation method for helicopters based on the Chinese Beidou position and attitude state vector, belonging to the technical field of helicopter navigation, including: obtaining helicopter information based on Beidou satellites, converting the helicopter information to generate a high-precision position and attitude information vector; obtaining the coordinates of the helicopter, importing the coordinates into three-dimensional map data to generate grid map target data; constructing a perception model, processing the high-precision position and attitude information vector to obtain helicopter motion and map feature data; constructing an action calculation model based on a deep Q network, calculating the data based on the action calculation model to obtain the Q value of each action of the helicopter; performing path planning based on the Q value of each action of the helicopter to generate a short baseline navigation path plan. In this application, the Chinese Beidou satellites have high-precision position and attitude information, which helps the helicopter to achieve a safe switch of attitude changes on a smaller path and is more suitable for various maneuvering scenario applications.

Description

Technical Field

[0001] The present invention belongs to the technical field of helicopter navigation, and particularly relates to a helicopter short baseline navigation method based on the Beidou position and attitude state vector in China. Background Technique

[0002] The Beidou satellite navigation system is a global satellite positioning and communication system independently developed by China, and it is the third mature satellite navigation system after the US Global Positioning System and the Russian Global Navigation Satellite System. The Beidou satellite navigation system consists of a space segment, a ground segment, and a user segment. It can provide high-precision and highly reliable positioning, navigation, and timing services for various users all-weather and all-time globally, and has the ability of short message communication, providing passive positioning, navigation, and timing services for the Asia-Pacific region. The Beidou satellite navigation system has passed the recognition of the International Maritime Organization and has become an integral part of the global radio navigation system. It is a recognized supplier by the United Nations Committee on the Peaceful Uses of Outer Space. The Beidou satellites in China have high-precision position and attitude information, which helps helicopters to achieve safe switching of attitude changes on a smaller path and is more suitable for various maneuvering scenarios.

[0003] During the flight of a helicopter, due to the operation of the rotor and engine, strong vibrations will be generated. Such vibrations can significantly affect the performance of on-board sensors (such as inertial measurement units IMU, gyroscopes, and accelerometers), resulting in an increase in data noise, thereby reducing the accuracy and reliability of the navigation system. The flight dynamics of helicopters are much more complex than those of fixed-wing aircraft. They can perform vertical takeoff and landing, hover, and quickly change the flight direction. Such a high degree of maneuverability requires the navigation system to be able to accurately reflect the position and attitude changes in real time. Therefore, there is an urgent need for a helicopter short baseline navigation method based on the Beidou position and attitude state vector in China to solve the above problems. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a helicopter short baseline navigation method based on the Beidou position and attitude state vector in China to solve the problems existing in the above prior art.

[0005] To achieve the above object, the present invention provides a helicopter short baseline navigation method based on the Beidou position and attitude state vector in China, including:

[0006] Obtaining the position, speed, and time information of the helicopter based on Beidou satellites, and transforming the position, speed, and time information of the helicopter to generate a high-precision position and attitude information vector;

[0007] Obtaining the starting point coordinates and ending point coordinates of the helicopter, importing the starting point coordinates and ending point coordinates into the three-dimensional map data and dividing the grid to generate grid map target data;

[0008] Construct a perception model to process the high-precision pose information vector and the grid map target data to obtain helicopter action feature data and map feature data;

[0009] Construct an action calculation model based on the deep Q-network, and calculate the helicopter action feature data and the map feature data based on the action calculation model to obtain the Q value of each action of the helicopter;

[0010] Perform path planning based on the Q value of each action of the helicopter to generate a navigation result.

[0011] Preferably, the process of generating the high-precision pose information vector includes:

[0012] Obtain the operating parameters of the helicopter, and at the same time capture the signals of multiple Beidou satellites through the receiver of the helicopter to generate the position, speed and time information of the helicopter;

[0013] Obtain the attitude information of the helicopter based on an electronic compass and an inertial navigation device, and the attitude information includes roll angle information, pitch angle information and yaw angle information;

[0014] Fuse the attitude information and the position vector to generate a pose vector;

[0015] Annotate the time information as an additional element to the pose vector to generate the high-precision pose information vector.

[0016] Preferably, the process of obtaining the starting point coordinates and the ending point coordinates of the helicopter, importing the starting point coordinates and the ending point coordinates into the three-dimensional map data and dividing the grid to generate the grid map target data includes:

[0017] Obtain the starting point coordinates and the ending point coordinates of the helicopter, and convert the starting point coordinates and the ending point coordinates of the helicopter into three-dimensional coordinates;

[0018] Obtain the spatial data within the range of the starting point coordinates and the ending point coordinates of the helicopter, and convert the spatial data into three-dimensional map data;

[0019] Divide the three-dimensional map data into grid networks;

[0020] Import the three-dimensional coordinates into the grid network to obtain the grid map target data.

[0021] Preferably, the process of constructing the perception model to process the high-precision pose information vector and the grid map target data to obtain helicopter action feature data and map feature data includes:

[0022] Extract the route shape features from the grid map target data through the dense layer of the perception model;

[0023] Convert the high-precision pose information vector into a parameter matrix, and obtain the preliminary operation strategy of the helicopter through the parameter matrix;

[0024] Extract features from the preliminary operation strategy through the perception model to obtain the helicopter state vector information.

[0025] Preferably, the perception model is a network model based on the YOLOv5 model, and a spatial attention mechanism module is inserted between the C2f module and the Conv module of the backbone network of the YOLOv5 model;

[0026] The spatial attention mechanism module is a SENet module.

[0027] Preferably, the process of constructing the action calculation model based on the deep Q network includes:

[0028] Obtain the state representation of the helicopter, and the state representation includes the position, speed and attitude of the helicopter;

[0029] Define the available action space of the helicopter;

[0030] Construct a deep Q network model based on a fully connected neural network and the available action space of the helicopter, use the state representation as the input of the deep Q network model, and the output is the Q value of each action;

[0031] Construct the loss function of the deep Q network model, and the loss function is the objective function in the Q-learning algorithm;

[0032] Train the deep Q network model based on experience replay and target network. The deep Q network model learns the optimal action strategy through interaction with the environment to maximize the long-term cumulative reward and generate the action calculation model.

[0033] Preferably, the process of calculating the Q value of each action of the helicopter based on the action calculation model for the helicopter action feature data and the map feature data includes:

[0034] Calculate the helicopter action feature data and the map feature data through the action calculation model to obtain the helicopter action Q value;

[0035] Convert the Q value of each action of the helicopter into the corresponding action and simulate and execute it through the device to generate a simulated action;

[0036] Import the simulated action into the surrounding environment morphological features to update the environment state and generate an environment feedback;

[0037] Update the Q value of each device based on the environmental feedback to generate the Q value of each action of the helicopter.

[0038] Preferably, the process of generating the short baseline navigation path planning result based on the Q value of each action of the helicopter includes:

[0039] Generate instructions and waypoints from the Q value of each action of the helicopter;

[0040] Perform path planning for the helicopter based on the instructions and waypoints to generate the navigation result.

[0041] Compared with the prior art, the present invention has the following advantages and technical effects:

[0042] In this application, the Beidou satellite has high positioning accuracy and can provide positioning accuracy at the meter level or even sub-meter level, which helps the helicopter to achieve precise path planning and navigation. The accuracy and reliability of the navigation system of the helicopter are constructed through the Beidou satellite. At the same time, the path planning can accurately reflect the pose change in real time. And the deep Q algorithm in the present invention has strong adaptability and can perform adaptive learning according to the changes and complexity of the environment, so as to adapt to different flight scenarios and mission requirements. And the deep Q algorithm can learn the optimal action strategy of the helicopter in different states from a large amount of experience, and can adjust the path planning in real time according to the environmental feedback to maximize the long-term cumulative reward. Description of the Drawings

[0043] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0044] Figure 1 It is a flowchart of the short baseline navigation method for a helicopter based on the Beidou position and pose state vector in the embodiment of the present invention. Detailed Embodiments

[0045] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.

[0046] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions. And, although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0047] Embodiment 1

[0048] As Figure 1As shown in the figure, in this embodiment, a helicopter short baseline navigation method based on the Chinese Beidou position and attitude state vector is provided, including:

[0049] Obtain the position, attitude and time information of the helicopter based on Beidou satellites, helicopter electronic compasses and inertial navigation equipment, and fuse the position, attitude and time information of the helicopter to generate a high-precision position and attitude information vector;

[0050] Obtain the starting point coordinates and ending point coordinates of the helicopter, import the starting point coordinates and ending point coordinates into the three-dimensional map data and divide the grid to generate grid map target data;

[0051] Construct a perception model, process the high-precision position and attitude information vector and the grid map target data to obtain helicopter action feature data and map feature data;

[0052] Construct an action calculation model based on the deep Q network, calculate the helicopter action feature data and the map feature data based on the action calculation model, and obtain the Q value of each action of the helicopter;

[0053] Perform path planning based on the Q value of each action of the helicopter to generate a short baseline navigation path planning result.

[0054] Obtain the position, attitude and time information of the helicopter based on Beidou satellites, helicopter electronic compasses and inertial navigation equipment, and fuse the position, attitude and time information of the helicopter to generate a high-precision position and attitude information vector;

[0055] The Beidou system provides global positioning services through a group of satellites. The signals emitted by these satellites are captured by the receivers on the helicopter. The helicopter receivers calculate the signal propagation time from the satellites to the receivers, and then estimate the distance between the helicopter and the satellites. Using the data of at least four satellites, the receivers of the helicopter can calculate the three-dimensional position (longitude, latitude, altitude) and time information.

[0056] The electronic compass is used to measure the direction of the helicopter relative to the Earth's magnetic north pole. A magnetic sensor (such as a Hall effect sensor) is used to detect the direction and intensity of the Earth's magnetic field.

[0057] The inertial navigation system uses accelerometers and gyroscopes to measure and calculate the speed, direction and attitude of the aircraft.

[0058] The accelerometer measures the acceleration of the helicopter in three-dimensional space, while the gyroscope measures its rotation rate.

[0059] INS can work independently without external reference, so it is particularly important when GPS signals are unavailable.

[0060] Obtain the starting coordinates and ending coordinates of the helicopter, import the starting coordinates and ending coordinates into the 3D map data and divide the grid to generate grid map target data;

[0061] Construct a perception model, process the high-precision pose information vector and the grid map target data to obtain helicopter action feature data and map feature data;

[0062] Based on the deep Q-network, construct an action calculation model, and calculate the helicopter action feature data and the map feature data based on the action calculation model to obtain the Q value of each action of the helicopter;

[0063] Based on the Q value of each action of the helicopter, perform path planning to generate a short baseline navigation path planning result.

[0064] For a further optimized solution, the process of generating the high-precision pose information vector includes:

[0065] Obtain the operating parameters of the helicopter, and at the same time capture the signals of multiple Beidou satellites through the receiver of the helicopter to generate the position, speed and time information of the helicopter;

[0066] The helicopter is equipped with receivers for receiving Beidou satellite signals. These receivers can capture the signals from multiple Beidou satellites to determine the position, speed and time information of the helicopter.

[0067] The received satellite signals are processed through positioning and resolution algorithms to determine the accurate position of the helicopter.

[0068] Convert the position information of the helicopter into a 3D vector to generate the position vector of the helicopter;

[0069] Convert the speed information of the helicopter into a speed vector;

[0070] Annotate the time information as an additional element to the position vector and the speed vector to generate the high-precision pose information vector.

[0071] The position of the helicopter can be represented by a point in a 3D coordinate system, which includes longitude, latitude and altitude. This point can be converted into a 3D vector, where each component represents the corresponding coordinate value.

[0072] The speed of the helicopter can be represented by a speed vector, including horizontal speed and vertical speed. The horizontal speed can be represented by the speed components in the longitude and latitude directions, while the vertical speed can be represented by the speed component in the altitude direction.

[0073] Time information is usually not directly converted into vectors. However, in navigation and motion equations, time plays an important role in calculating the position and speed of a helicopter at different time points.

[0074] Add an extra element to the state vector to represent the current time. This timestamp can be absolute time, such as UTC time or local time, or it can be time relative to a certain starting time point. For example, the state vector may include a position vector, a speed vector, and an element representing the current time.

[0075] For a further optimized solution, the process of obtaining the starting coordinates and ending coordinates of the helicopter, importing the starting coordinates and ending coordinates into the 3D map data and dividing the grid to generate grid map target data includes:

[0076] Obtain the starting coordinates and ending coordinates of the helicopter, and convert the starting coordinates and ending coordinates of the helicopter into 3D coordinates;

[0077] Obtain the spatial data within the range of the starting coordinates and ending coordinates of the helicopter, and convert the spatial data into 3D map data;

[0078] Divide the 3D map data into grid networks;

[0079] Import the 3D coordinates into the grid network to obtain the grid map target data.

[0080] For a further optimized solution, the process of constructing the perception model, processing the high-precision pose information vector and the grid map target data to obtain helicopter action feature data and map feature data includes:

[0081] Extract the grid map target data through the dense layer of the perception model to obtain the route form features;

[0082] Convert the high-precision pose information vector into a parameter matrix, and obtain the preliminary operation strategy of the helicopter through the parameter matrix;

[0083] Extract features from the preliminary operation strategy through the perception model to obtain the helicopter state vector information.

[0084] For a further optimized solution, the perception model is a network model based on the YOLOv5 model, and a spatial attention mechanism module is inserted between the C2f module and the Conv module of the backbone network of the YOLOv5 model;

[0085] The spatial attention mechanism is a mechanism used to enhance the neural network's modeling ability in the spatial dimension. It enables the network to better understand and utilize the spatial context information by learning the relationships between different positions in the feature map.

[0086] First, determine the position to insert the spatial attention mechanism between the C2f module and the Conv module. Usually, such a mechanism is inserted in the shallow feature extraction stage of the backbone network to utilize richer spatial information.

[0087] Select a suitable spatial attention module. Common choices include the self-attention mechanism (Self-Attention), SENet (Squeeze-and-Excitation Networks), etc. These modules can effectively capture the relationships between different positions in the feature map and dynamically adjust the weights of the feature map.

[0088] At the determined position, insert the selected spatial attention module into the backbone network. This usually involves adding new network layers between specific layers to implement the calculation and feature fusion of the spatial attention mechanism.

[0089] Fine-tune and train the modified model so that the spatial attention mechanism can effectively learn the feature representation suitable for the task. During the training process, optimize the model parameters through the backpropagation algorithm to minimize the loss function.

[0090] The spatial attention mechanism module is the SENet module.

[0091] For the further optimization scheme, the process of constructing the action calculation model based on the deep Q-network includes:

[0092] Obtain the state representation of the helicopter, and the state representation includes the position, speed, and attitude of the helicopter;

[0093] Define the available action space of the helicopter;

[0094] Build a deep Q-network model based on a fully connected neural network and the available action space of the helicopter, use the state representation as the input of the deep Q-network model, and the output is the Q value of each action;

[0095] Construct the loss function of the deep Q-network model, and the loss function is the objective function in the Q-learning algorithm;

[0096] In the Q-learning algorithm, the objective function is usually represented by the Bellman Equation. The Bellman Equation describes the relationship between the Q value of the current state and the Q value of the next state, and is the core of updating the Q value in the Q-learning algorithm. Assume that the current state is s, and after taking action a, it enters the next state's', then the Bellman Equation can be expressed as:

[0097] [Q(s,a) = r+\gamma\max_{a'}Q(s',a')]

[0098] where:

[0099] (Q(s,a)) represents the Q-value of taking action (a) in state (s);

[0100] (r) represents the immediate reward obtained after taking action (a) in state (s);

[0101] (\gamma) represents the discount factor, which is used to measure the importance of future rewards and usually takes values between [0, 1];

[0102] (s') represents the next state entered;

[0103] (a') represents all possible actions in the next state (s');

[0104] (\max_{a'}Q(s',a')) represents the maximum Q-value after taking all possible actions in the next state (s').

[0105] The meaning of the objective function is that the Q-value of taking action (a) in the current state (s) is equal to the current immediate reward (r) plus the maximum Q-value of the next state (s'), multiplied by the discount factor (\gamma). This represents a method of discounting future rewards to the current state to guide the agent to consider long-term future benefits when making decisions. In the Q-learning algorithm, the objective function is used to update the Q-value to approximate the right side of the Bellman equation.

[0106] The deep Q-network model is trained based on experience replay and a target network. The deep Q-network model learns the optimal action policy through interaction with the environment to maximize the long-term cumulative reward and generates the action calculation model.

[0107] First, the problem environment is modeled as a Markov decision process (MDP), including a state space, an action space, a reward function, and a state transition probability.

[0108] The goal of the deep Q-network is to approximately calculate the action value function (Q-value), that is, the expected value of the cumulative reward obtained after taking a certain action in a given state. A neural network is used to represent the Q-value function, with the state as the input and the Q-values of each possible action as the output.

[0109] State and action representation: The state in the environment is represented as the input of the neural network, and a Q-value is output for each action. Therefore, the input dimension of the neural network is the dimension of the state, and the output dimension is the number of actions.

[0110] The training objective of the Deep Q-Network is to minimize the mean squared error (MSE) between the current Q-value estimate and the target Q-value. The target Q-value is calculated based on the Bellman equation and represents the expected cumulative reward after choosing the optimal action in the current state.

[0111] To improve the utilization efficiency and stability of samples, the Deep Q-Network introduces an experience replay mechanism. During training, the state transition samples at each moment are stored in the experience replay buffer, and a batch of samples is randomly sampled for training the neural network.

[0112] To improve the stability of training, the Deep Q-Network also introduces a Fixed Target Network mechanism. At each training update, the parameters of the current Q-value network are copied to a target Q-value network for calculating the target Q-value, instead of directly using the current Q-value network.

[0113] In each training step, a batch of samples is sampled from the experience replay buffer, the Q-value of the current Q-value network and the target Q-value of the target Q-value network are calculated, and then the mean squared error is used to update the parameters of the current Q-value network. This process is repeated until the stopping condition is reached.

[0114] The role of outputting the Q-values of each possible action is to help the agent select the optimal action in a given state. In reinforcement learning, the Q-value represents the expected cumulative reward that can be obtained after choosing a certain action in the current state. Therefore, the agent can choose the action that is most likely to obtain the maximum reward by comparing the Q-values of each action. By continuously updating the Q-value function, the agent can learn which action should be chosen in different states, thus achieving autonomous decision-making and behavior.

[0115] In the Deep Q-Network, each device has a corresponding Q-value output. By taking the environmental state and the state of each device as inputs, the Q-values of each device are calculated respectively.

[0116] For each device, the optimal action is selected according to its corresponding Q-value. It can simply select the action with the highest Q-value as the next action, or select the action according to a certain exploration strategy to better explore the environment and improve the learning efficiency.

[0117] The selected action is applied to each device, enabling it to execute the corresponding action and update the environmental state.

[0118] According to the feedback from the environment, including rewards or punishments, the Q-values of each device are updated so as to learn a better strategy in subsequent training.

[0119] In this way, the Q-value of the next action can be calculated for each device separately, and the optimal action can be selected based on the Q-value. Such a method enables the agent to better adapt to the characteristics of different devices and environmental changes, thus achieving more effective path planning and decision-making.

[0120] For a further optimized solution, the process of calculating the Q-value of each action of the helicopter based on the action calculation model for the helicopter action feature data and the map feature data includes:

[0121] Calculating the helicopter action Q-value by calculating the helicopter action feature data and the map feature data through the action calculation model;

[0122] Converting the Q-value of each action of the helicopter into the corresponding action and simulating the execution through the device to generate a simulated action;

[0123] Importing the simulated action into the morphological features of the surrounding environment, updating the environmental state, and generating an environmental feedback;

[0124] Updating the Q-value of each device based on the environmental feedback to generate the Q-value of each action of the helicopter.

[0125] For a further optimized solution, the process of generating a navigation result by performing path planning based on the Q-value of each action of the helicopter includes:

[0126] Generating instructions and path points from the Q-value of each action of the helicopter;

[0127] Performing path planning on the helicopter based on the instructions and path points to generate the navigation result.

[0128] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A helicopter short-baseline navigation method based on the Chinese Beidou position and attitude state vector, characterized in that, The method includes the following steps: Based on Beidou satellites, a helicopter electronic compass, and inertial navigation equipment, obtain the position, attitude, and time information of the helicopter, and fuse the position, attitude, and time information of the helicopter to generate a high-precision pose information vector; Obtain the starting coordinate and ending coordinate of the helicopter, import the starting coordinate and ending coordinate into three-dimensional map data and divide the grid to generate grid map target data; Construct a perception model, process the high-precision pose information vector and the grid map target data to obtain helicopter action feature data and map feature data; Construct an action calculation model based on the deep Q network, calculate the helicopter action feature data and the map feature data based on the action calculation model to obtain the Q value of each action of the helicopter; Based on the Q value of each action of the helicopter, perform path planning to generate a short baseline navigation path planning result.

2. The helicopter short baseline navigation method based on the Chinese Beidou position and attitude state vector according to claim 1, wherein, The process of generating the high-precision pose information includes: Obtain the operating parameters of the helicopter, and at the same time capture the signals of multiple Beidou satellites through the receiver of the helicopter to generate the position, speed, and time information of the helicopter; Convert the position information of the helicopter into a three-dimensional vector to generate a position vector of the helicopter; Obtain the attitude information of the helicopter based on the electronic compass and inertial navigation equipment, and the attitude information includes roll angle information, pitch angle information, and yaw angle information; Fuse the attitude information and the position vector to generate a pose vector; Annotate the time information as an additional element to the pose vector to generate the high-precision pose information vector.

3. The helicopter short-baseline navigation method based on the Chinese Beidou position and attitude state vector according to claim 1, characterized in that, The process of obtaining the starting coordinate and ending coordinate of the helicopter, importing the starting coordinate and ending coordinate into three-dimensional map data and dividing the grid to generate grid map target data includes: Obtain the starting coordinate and ending coordinate of the helicopter, and convert the starting coordinate and ending coordinate of the helicopter into three-dimensional coordinates; Obtain the spatial data within the range of the starting coordinate and ending coordinate of the helicopter, and convert the spatial data into three-dimensional map data; Divide the three-dimensional map data into grid networks; Import the three-dimensional coordinates into the grid network to obtain the grid map target data.

4. The helicopter short baseline navigation method based on the Chinese Beidou position and attitude state vector according to claim 1, characterized in that The process of constructing the perception model, processing the high-precision pose information vector and the grid map target data to obtain helicopter action feature data and map feature data includes: Extract the route shape features from the grid map target data through the dense layer of the perception model; Convert the high-precision pose information vector into a parameter matrix, and obtain the preliminary operation strategy of the helicopter through the parameter matrix; Extract features from the preliminary operation strategy through the perception model to obtain the helicopter action feature data and map feature data.

5. The helicopter short-baseline navigation method based on the Chinese Beidou position and attitude state vector according to claim 1, wherein The perception model is a network model based on the YOLOv5 model, and a spatial attention mechanism module is inserted between the C2f module and the Conv module of the backbone network of the YOLOv5 model; The spatial attention mechanism module is a SENet module.

6. The helicopter short baseline navigation method based on the Chinese Beidou position and attitude state vector according to claim 5, wherein, The process of constructing the action calculation model based on the deep Q network includes: Obtain the state representation of the helicopter, where the state representation includes the position, speed, and attitude of the helicopter; Define the action space available to the helicopter; Construct a deep Q-network model based on a fully connected neural network and the action space available to the helicopter, use the state representation as the input of the deep Q-network model, and the output is the Q-value for each action; Construct the loss function of the deep Q-network model, where the loss function is the objective function in the Q-learning algorithm; Train the deep Q-network model based on experience replay and a target network. The deep Q-network model learns the optimal action policy through interaction with the environment to maximize the long-term cumulative reward and generate the action calculation model.

7. The helicopter short-baseline navigation method based on the Chinese Beidou position and attitude state vector according to claim 1, characterized in that The process of obtaining the Q-value for each action of the helicopter based on the action calculation model for the helicopter action feature data and map feature data includes: Calculate the helicopter action feature data and map feature data through the action calculation model to obtain the helicopter action Q-value; Convert the Q-value for each action of the helicopter into the corresponding action and simulate the execution through the device to generate a simulated action; Import the simulated action into the surrounding environmental morphological features to update the environmental state and generate an environmental feedback; Update the Q-value of each device based on the environmental feedback to generate the Q-value for each action of the helicopter.

8. The helicopter short-baseline navigation method based on the Chinese Beidou position and attitude state vector according to claim 1, wherein The process of generating a short baseline navigation path planning result based on the Q-value for each action of the helicopter includes: Generate short baseline navigation instructions and path points within 10 meters from the Q-value for each action of the helicopter; Perform path planning for the helicopter based on the instructions and path points to generate navigation path switching information.

Images