An adaptive unmanned aerial vehicle intelligent navigation method and system

Abstract

The present application relates to the field of unmanned aerial vehicle intelligent navigation, and aims to solve the problem that the sensing angle of the unmanned aerial vehicle cannot be changed and the uncertainty information cannot be actively acquired and eliminated, and provides an adaptive unmanned aerial vehicle intelligent navigation method and system.The adaptive unmanned aerial vehicle intelligent navigation method comprises training a basic navigation model, calculating the perplexity of the generated navigation path points, and marking the navigation path points with perplexity higher than a set threshold as high perplexity path points; a height adjustment prediction module is added to the basic navigation model, and the basic navigation model after the addition of the height adjustment prediction module is trained again by using the marked high perplexity path points; wherein the height adjustment prediction module determines whether to trigger height adjustment according to whether the navigation path point is a high perplexity path point; the trained basic navigation model is used to automatically trigger height adjustment when decision making is difficult, so that the unmanned aerial vehicle actively changes the flight height in a complex environment to acquire observation information with the optimal angle of view, and thus the most reliable decision can be made.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of unmanned aerial vehicle intelligent navigation, and in particular to an adaptive unmanned aerial vehicle intelligent navigation method and system. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] The goal of unmanned aerial vehicle visual language navigation is to enable the unmanned aerial vehicle to navigate autonomously to a target location in an unknown environment according to natural language instructions combined with its own visual perception. Its application prospects are wide, such as intelligent inspection, disaster search and rescue, and human-computer interaction. However, the existing technology still faces many challenges in realizing unmanned aerial vehicle visual language navigation. One of the core challenges is how to effectively handle the uncertainty in the decision-making process. When the navigation model faces instructions with ambiguous semantics or environments with visual occlusion, there may be high uncertainty about the action to be performed next. This uncertainty will directly affect the efficiency and safety of the unmanned aerial vehicle in completing the task.

[0004] Existing research usually constructs a navigation agent with dialogue capability based on rules or neural networks, allowing the unmanned aerial vehicle to seek external guidance when it is difficult to make a decision. However, this method mainly focuses on semantic-level disambiguation processing and fails to fundamentally change the perception perspective of the unmanned aerial vehicle and actively obtain new information that can eliminate uncertainty. There is still a lack of mechanisms that can adaptively use high variability to optimize visual information acquisition. SUMMARY

[0005] To solve the above technical problems, the present application provides an adaptive unmanned aerial vehicle intelligent navigation method and system, which can enable the unmanned aerial vehicle to actively and adaptively adjust its flight altitude when the decision-making uncertainty is high, obtain visual information at the most favorable perspective, and make the most reliable decision.

[0006] To achieve the above purpose, the present application adopts the following technical solutions:

[0007] The first aspect of the present application provides an adaptive unmanned aerial vehicle intelligent navigation method.

[0008] In one or more embodiments, an adaptive unmanned aerial vehicle intelligent navigation method is provided, comprising:

[0009] training a basic navigation model, calculating the perplexity of the generated navigation path points, and marking the navigation path points with perplexity higher than a set threshold as high perplexity path points; wherein the perplexity is calculated based on the weighted calculation of action entropy and environment semantic matching degree;

[0010] An altitude adjustment prediction module is added to the basic navigation model, and the basic navigation model with the added altitude adjustment prediction module is retrained using marked high confusion waypoints; the altitude adjustment prediction module determines whether to trigger altitude adjustment based on whether the navigation waypoint is a high confusion waypoint.

[0011] By utilizing a well-trained basic navigation model, altitude adjustments can be automatically triggered when decision-making is difficult, enabling the UAV to proactively change its flight altitude in complex environments in order to obtain observation information from the optimal perspective.

[0012] As one implementation method, the representation function of perplexity is:

[0013] ;

[0014] in, express The level of confusion at any given moment; and There are two hyperparameters; It is a moment The degree of semantic matching in the context of the environment. Indicates time The entropy of the action.

[0015] As one implementation method, action entropy is calculated from the action probability distribution output by the model. The larger the action entropy, the greater the uncertainty in the decision. The calculation formula is:

[0016] ;

[0017] in, Indicates time The entropy of action, Represents the set of all navigation actions. Indicates the model at time... Predict the execution action The probability of.

[0018] As one implementation method, the environmental semantic matching degree analyzes the observed image using a target detection model to calculate the average confidence score of targets related to language instructions. The calculation formula is as follows:

[0019] ;

[0020] in, It is a moment The degree of semantic matching in the context of the environment. This indicates the target detection model's detection of objects within the current field of view that match the instructions. The number of relevant targets It is the first The confidence score of each detected target.

[0021] As an implementation, the total loss function of the base navigation model after adding the height adjustment prediction module is set as , which is expressed as:

[0022] ;

[0023] ;

[0024] ;

[0025] wherein, is the loss function in the navigation process; is the height decision loss function; represents the probability that the model predicts the execution of the expert action ; represents the total number of path points, represents the true label of the path point under the category , if the path point should trigger height adjustment, then , otherwise ; is the predicted height adjustment probability distribution of the category at the path point , represents the probability of not needing to adjust the height, represents the probability of predicting the need to adjust the height.

[0026] As an implementation, in the process of automatically triggering height adjustment when decision-making is difficult by using the trained base navigation model, when the final target is detected in the field of view of the UAV, in order to obtain better observation information, the UAV should approach the target height, and the new height of the UAV is expressed as:

[0027] ;

[0028] wherein, represents the axis coordinate value, i.e., the height value, of the target object in the world coordinate system.

[0029] As an implementation, in the process of automatically triggering height adjustment when decision-making is difficult by using the trained base navigation model, when the target is not detected, in order to expand the search range, the UAV performs an ascending operation, and the new height of the UAV is adaptively determined according to the environmental conditions as:

[0030] ;

[0031] wherein, Indicates the drone's current altitude; Represents the set of other observed objects; Represents objects Height; , It is a constant; This represents the maximum value function.

[0032] A second aspect of the present invention provides an adaptive intelligent navigation system for unmanned aerial vehicles (UAVs).

[0033] In one or more embodiments, an adaptive unmanned aerial vehicle (UAV) intelligent navigation system includes:

[0034] The high-confusion pathpoint marking module is used to train the base navigation model. It calculates the confusion degree of the generated navigation pathpoints and marks navigation pathpoints with confusion degrees higher than a set threshold as high-confusion pathpoints. The confusion degree is calculated based on a weighted average of action entropy and environmental semantic matching degree.

[0035] The basic navigation model retraining module is used to add an altitude adjustment prediction module to the basic navigation model. It retrains the basic navigation model after adding the altitude adjustment prediction module using marked high confusion waypoints. The altitude adjustment prediction module determines whether to trigger altitude adjustment based on whether the navigation waypoint is a high confusion waypoint.

[0036] The adaptive altitude adjustment module is used to automatically trigger altitude adjustment when decision-making is difficult by utilizing a trained basic navigation model, enabling the UAV to actively change its flight altitude in complex environments in order to obtain observation information from the optimal perspective.

[0037] A third aspect of the present invention provides a computer-readable storage medium.

[0038] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the adaptive unmanned aerial vehicle intelligent navigation method described above.

[0039] A fourth aspect of the present invention provides an electronic device.

[0040] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the adaptive unmanned aerial vehicle intelligent navigation method described above.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] This invention innovatively introduces confusion level to mark navigation waypoints. By combining the confusion level index with action entropy and environmental semantic matching degree, it accurately models and evaluates the decision-making uncertainty of UAVs during complex navigation processes in real time. When confused, the UAV dynamically adjusts its flight altitude to proactively acquire optimal visual perception information. By adjusting the altitude, it can not only overcome local occlusion and expand the field of view, but also achieve multi-level perception of the environment at different scales, thus providing more comprehensive and reliable input for subsequent path planning and decision-making. Attached Figure Description

[0043] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0044] Figure 1 This is a flowchart illustrating the adaptive unmanned aerial vehicle (UAV) intelligent navigation method according to an embodiment of the present invention.

[0045] Figure 2 This is a schematic diagram of the adaptive unmanned aerial vehicle (UAV) intelligent navigation system according to an embodiment of the present invention;

[0046] Figure 3 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0047] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0048] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0049] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0050] Unlike ground robots, which are limited to two-dimensional planar movement, drones possess the natural advantage of free movement along the vertical dimension. Flight altitude is crucial for the visual perception of drones. At higher altitudes, drones can obtain a macroscopic bird's-eye view, which helps in understanding the overall topology of the scene, thereby reducing uncertainty in global path planning. At lower altitudes, drones can acquire high-resolution local details, which is beneficial for identifying small markers on the ground or confirming the specific features of targets. Current research still lacks mechanisms that can adaptively utilize altitude changes to optimize visual information acquisition.

[0051] To address the decision uncertainty problem in UAV visual language navigation, this invention proposes an adaptive UAV intelligent navigation method. By introducing a "confusion degree" index to identify and learn when the UAV is making difficult decisions, the UAV can proactively change its flight altitude in complex environments to obtain observation information from the optimal perspective, effectively reducing decision uncertainty.

[0052] The goal of drone visual language navigation is to learn a strategy. This enables the drone intelligent agent to respond to natural language commands. Perform a series of actions in an unknown three-dimensional environment. Ultimately, it reaches the target location described in the instruction. Specifically, at each time step... The agent receives current visual observations, including RGB images from a first-person perspective. and depth images And based on these observations and natural language instructions To select the next action, that is .

[0053] Figure 1 This is a flowchart illustrating an adaptive unmanned aerial vehicle (UAV) intelligent navigation method according to an embodiment of the present invention, as shown below. Figure 1 The adaptive UAV intelligent navigation method shown in this embodiment may include the following steps S101 to S103.

[0054] The specific implementation process of steps S101 to S103 is as follows:

[0055] Step S101: Train the basic navigation model, calculate the perplexity of the generated navigation path points, and mark navigation path points with perplexity higher than a set threshold as high perplexity path points; wherein, the perplexity is calculated based on the weighted sum of action entropy and environmental semantic matching degree.

[0056] The basic navigation model adopts a multimodal fusion architecture, including a visual encoder. Depth image encoder Language encoder Multimodal fusion module and action decision head .

[0057] The visual encoder is responsible for processing the input image Extracting high-dimensional visual features Typically, pre-trained models such as ResNet are used, and this process can be represented as:

[0058] ;

[0059] Depth image encoder extracts depth image Feature representation This typically involves using pre-trained models such as ResNet, and the process can be represented as follows:

[0060] The language encoder is responsible for extracting natural language instructions. Feature representation Typically, pre-trained models such as BERT are used, and this process can be represented as:

[0061] ;

[0062] Subsequently, the three types of feature vectors are concatenated along the channel dimension to construct a unified multimodal fusion input vector:

[0063] ;

[0064] in This indicates a feature concatenation operation. This represents the spliced ​​multimodal features.

[0065] The above features are input into the multimodal fusion module. This module fuses and infers multimodal information to ultimately generate feature representations of the next waypoint. :

[0066] ;

[0067] In one alternative embodiment, the multimodal fusion module A GRU architecture is adopted. To fully utilize the sequence modeling capabilities of GRU, sequences from different time points are... The sequence formed is used as input:

[0068]

[0069] GRU iteratively updates the input sequence through a gating mechanism:

[0070]

[0071]

[0072]

[0073] in, Represents the Sigmoid function; Represents the hyperbolic tangent activation function; It represents the Hadamardi (or Hadama) stack; , , , and Represents the learnable weight matrix; This indicates resetting the door and controlling its previous hidden state. The proportion of forgotten information in the content; This indicates an update gate, determining the new candidate state. Compared to the previous hidden state The fusion ratio; The candidate hidden state is determined by combining the current input with the historical state after being filtered by the reset gate. The hidden state at the current moment is the weighted combination of the previous hidden state and the candidate state. The hidden state at the end of the sequence, representing all temporal visual, depth, and linguistic information, is used as the final feature table for waypoints: .

[0074] Finally, the action decision head is utilized. The waypoint features are mapped to a predefined navigation action space, and the probability distribution of the actions is output. :

[0075] ;

[0076] in, This represents the softmax operation, which transforms any real vector into a probability distribution that sums to 1.

[0077] To enable the model to learn the correct navigation strategy, this embodiment of the invention employs an imitation learning method, training the model using an expert demonstration dataset. This is achieved by minimizing the distribution of actions predicted by the model. The actions actually performed by the expert in this state The cross-entropy loss between nodes drives the optimization of model parameters, thereby making its behavior approximate expert policies. The loss function during navigation... The definition is as follows:

[0078] ;

[0079] in The model predicts the actions of the expert. The probability of.

[0080] After the model initially converges, a "Confusion Score" (CS) is introduced to quantify the uncertainty of the decision. This score is calculated from the action entropy. semantic matching degree with environment It consists of two parts.

[0081] Action entropy is calculated from the action probability distribution output by the model. The higher the action entropy, the greater the uncertainty in the decision. The calculation formula is:

[0082] ;

[0083] in, Indicates time The entropy of action, Represents the set of all navigation actions. Indicates the model at time... Predict the execution action The probability of.

[0084] The environmental semantic matching degree analyzes the observed images using a target detection model (e.g., the Grounding DINO model) and calculates the average confidence score of targets related to language instructions. The calculation formula is as follows:

[0085] ;

[0086] in, It is a moment The degree of semantic matching in the context of the environment. This indicates the target detection model's detection of objects within the current field of view that match the instructions. The number of relevant targets It is the first The confidence score of each detected target.

[0087] The representation function of perplexity is:

[0088] ;

[0089] in, express The level of confusion at any given moment; and There are two hyperparameters; It is a moment The degree of semantic matching in the context of the environment. Indicates time The entropy of the action.

[0090] Step S102: Add an altitude adjustment prediction module to the basic navigation model, and retrain the basic navigation model after adding the altitude adjustment prediction module using the marked high confusion waypoints; wherein, the altitude adjustment prediction module determines whether to trigger altitude adjustment based on whether the navigation waypoint is a high confusion waypoint.

[0091] After calculating the confusion score for all navigation waypoints, the waypoints are sorted according to their confusion score values, and the set of waypoints with the top K% confusion scores is denoted as the confused waypoint set. These points represent scenarios where the model is most uncertain or most prone to making incorrect decisions. For Generate a binary height adjustment pseudo-label to indicate whether the pathpoint should trigger a height adjustment, as defined below:

[0092] ;

[0093] in, Indicates at the waypoint This should trigger an altitude adjustment, Representing path points No height adjustment is needed. These automatically generated pseudo-labels will serve as supervisory signals for the next stage of policy learning.

[0094] In this phase, the architecture of the basic navigation model is enhanced by introducing an altitude-adjusted prediction head. This is used to determine whether the drone needs to perform an altitude adjustment in its current state. The enhanced model is then retrained using the maximum perplexity pathpoints labeled in the first stage.

[0095]

[0096] in, Indicates the model at the path point The height-adjusted probability distribution of the predicted location. This indicates the probability that the height does not need to be adjusted. This indicates the probability that the altitude needs to be adjusted.

[0097] When the UAV is at a confusing waypoint, a supervisory signal is applied to the altitude adjustment prediction head, causing it to output "yes," thus triggering an altitude adjustment operation. When the UAV is at a non-confusing waypoint, the supervisory prediction head outputs "no," meaning it maintains normal navigation without triggering an altitude adjustment. The altitude decision loss function is defined as follows:

[0098] ;

[0099] in, Indicates the total number of path points. Representing path points In category The actual label below, if the path point An altitude adjustment should be triggered. ,otherwise . For at the path point Category The height adjustment probability distribution of the forecast. This indicates the probability that the height does not need to be adjusted. This indicates the probability that the altitude needs to be adjusted.

[0100] The total loss function of the base navigation model after adding the altitude adjustment prediction module is set as follows: Its expression is:

[0101] ;

[0102] in, The loss function during navigation; This is the high-decision loss function.

[0103] Step S103: Utilize the trained basic navigation model to automatically trigger altitude adjustment when decision-making is difficult, enabling the UAV to actively change its flight altitude in complex environments in order to obtain observation information from the optimal perspective.

[0104] In actual deployment and testing, the UAV first acquires observation results at its current location and generates a current state representation based on the mission instructions. Then, the system uses the altitude decision head to reason about this state to determine whether an altitude adjustment operation needs to be performed: if the altitude decision head outputs a result indicating that no altitude adjustment is needed, the UAV directly enters the standard navigation process, selects and executes the optimal action in the current state; if the altitude decision head outputs a result indicating that altitude adjustment is needed, the system enters the altitude adjustment mode.

[0105] In high-adjustment mode, a large language model (such as GPT-4o) is first invoked to process natural language navigation instructions. Semantic parsing is performed. This process extracts a set of semantic entity categories from the instructions that are relevant to the current task and can be perceived in the physical world:

[0106] ;

[0107] in, Represents a specific semantic category, This represents a large language model. Based on the extracted semantic entity categories, the open-vocabulary object detection model GroundingDINO is used to detect objects in the current visual observation and returns the corresponding two-dimensional detection bounding boxes of the objects.

[0108] ;

[0109] in, This is the set of detection boxes for all targets. The detection bounding box for each target. The definition is as follows:

[0110] ;

[0111] in, This indicates the coordinates of the bottom left corner of the bounding box. This represents the coordinates of the top-right corner of the bounding box. The geometric center pixel coordinates of the detection box are calculated as follows:

[0112] ;

[0113] By combining the coordinates of the detection box center with the depth map information, the spatial position of the corresponding target in the camera coordinate system can be calculated. :

[0114] ;

[0115] in, The camera is at and Focal length in direction, These are the coordinates of the camera's principal point. This represents the depth value of the center pixel in the depth image.

[0116] Transform the coordinates from the camera coordinate system to the world coordinate system:

[0117] ;

[0118] in, Represents the coordinates of a three-dimensional point in the transformed world coordinate system. ; Let represent the rotation matrix from the camera to the world coordinate system at time t; It is a translation vector, representing the position of the camera in the world coordinate system at time t.

[0119] In the process of automatically triggering altitude adjustments when decision-making is difficult using a trained basic navigation model, when the final target is detected in the UAV's field of view, in order to obtain better observation information, the UAV should move closer to the target altitude. The UAV's new altitude... Represented as:

[0120] ;

[0121] in, Represents the target object in the world coordinate system The axis coordinate value, i.e., the height value.

[0122] When no target is detected, the drone ascends to expand its search range, reaching a new altitude. Based on adaptive determination according to environmental conditions:

[0123] ;

[0124] in, Indicates the drone's current altitude; Represents the set of other observed objects; Represents objects Height; , For constants (e.g.) , Those skilled in the art can set it specifically according to the actual situation); This represents the maximum value function.

[0125] After completing the altitude adjustment, the UAV will acquire new observations at the new altitude and make new navigation decisions based on the updated state representation.

[0126] The present invention has the following advantages:

[0127] (1) Enabling UAVs to have proactive environmental perception capabilities: Existing technologies, when faced with decision-making difficulties, mainly rely on requesting supplementary information from external systems to resolve ambiguities. In contrast, this invention constructs a proactive perception framework. When the system detects that the model is in a state of high uncertainty, it does not rely on external intervention but actively performs physical maneuvers (such as adaptively adjusting flight altitude) to collect environmental visual information with stronger discriminative capabilities in real time, thereby achieving autonomous resolution of uncertainty. This architecture not only enhances the autonomy of the model but also significantly reduces dependence on external resources.

[0128] (2) Calculation and application of navigation confusion: An innovative quantitative index of "confusion" is introduced. This index combines action entropy and environmental semantic matching degree to accurately model and evaluate the decision uncertainty of UAVs in complex navigation processes in real time.

[0129] (3) Implementation of the adaptive altitude adjustment mechanism: An adaptive altitude adjustment mechanism is proposed, which allows the UAV to dynamically adjust its flight altitude when confused, thereby actively acquiring the best visual perception information. By adjusting the altitude, it can not only overcome local occlusion and expand the field of view, but also perform multi-level perception of the environment at different scales, thus providing more comprehensive and reliable input for subsequent path planning and decision-making.

[0130] The adaptive UAV intelligent navigation method provided in this invention has a higher average success rate, a higher average oracle success rate, and a lower average navigation error. This adaptive UAV intelligent navigation method is applicable to the following scenarios:

[0131] Intelligent Unmanned Aerial Vehicle System: This invention can be applied to consumer-grade and industrial-grade intelligent unmanned aerial vehicles, enabling them to understand human natural language commands and navigate autonomously in unknown environments. It is widely applicable to scenarios such as express delivery, agricultural inspection, power line inspection, and disaster search and rescue, improving the flexibility and autonomy of mission execution.

[0132] Smart Cities: This invention can be integrated into city-level unmanned inspection platforms and emergency response systems to assist in tasks such as urban infrastructure inspection, traffic management, accident response, and regional search and rescue. It is particularly suitable for complex urban environments that require rapid deployment and dynamic response, thereby improving the intelligence level and safety assurance capabilities of urban operations.

[0133] like Figure 2 As shown, the adaptive UAV intelligent navigation system provided in this embodiment of the invention can be implemented in software. The adaptive UAV intelligent navigation system includes the following software modules: a high confusion waypoint marking module 201, a basic navigation model retraining module 202, and an adaptive altitude adjustment module 203.

[0134] The functions of each software module in the adaptive UAV intelligent navigation system are described below:

[0135] The high-confusion path point marking module 201 is used to train the basic navigation model, calculate the confusion degree of the generated navigation path points, and mark navigation path points with confusion degrees higher than a set threshold as high-confusion path points; wherein, the confusion degree is obtained by weighted calculation based on the action entropy and the environmental semantic matching degree;

[0136] The basic navigation model retraining module 202 is used to add an altitude adjustment prediction module to the basic navigation model and retrain the basic navigation model after adding the altitude adjustment prediction module using marked high confusion waypoints; wherein, the altitude adjustment prediction module determines whether to trigger altitude adjustment based on whether the navigation waypoint is a high confusion waypoint.

[0137] The adaptive altitude adjustment module 203 is used to automatically trigger altitude adjustment when decision-making is difficult by utilizing a trained basic navigation model, so that the UAV can actively change its flight altitude in complex environments in order to obtain observation information from the optimal perspective.

[0138] It should be noted that each module in the adaptive UAV intelligent navigation system of this embodiment corresponds one-to-one with each step in the adaptive UAV intelligent navigation method in the above embodiment, and their specific implementation process is the same, so it will not be repeated here.

[0139] The structure of the electronic device according to an embodiment of the present invention will be described in detail below. Figure 3 This is a schematic diagram of the composition structure of an electronic device provided in an embodiment of the present invention. It can be understood that... Figure 3 The diagram shows only an exemplary structure of the electronic device, not the entire structure. Some or all of the structures shown may be implemented as needed.

[0140] The electronic device provided in this embodiment of the invention includes: at least one processor 301, a memory 302, a user interface 303, and at least one network interface 304. The various components in the adaptive UAV intelligent navigation system are coupled together via a bus system 305. It can be understood that the bus system 305 is used to realize the connection and communication between these components. In addition to a data bus, the bus system 305 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 3 The general designated all buses as Bus System 305.

[0141] The user interface 303 may include a monitor, keyboard, mouse, trackball, click wheel, buttons, touchpad, or touch screen.

[0142] It is understood that memory 302 can be volatile memory or non-volatile memory, or both. In this embodiment of the invention, memory 302 is capable of storing data to support the operation of the terminal. Examples of this data include any computer programs used to operate on the terminal, such as operating systems and applications. The operating system includes various system programs, such as framework layers, core library layers, driver layers, etc., used to implement various basic services and handle hardware-based tasks. Applications can include various applications.

[0143] In some embodiments, the adaptive UAV intelligent navigation system provided by the present invention can be implemented using a combination of hardware and software. For example, the adaptive UAV intelligent navigation system provided by the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the adaptive UAV intelligent navigation method provided by the present invention. For instance, the processor in the form of a hardware decoding processor can employ one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0144] As an example, processor 301 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0145] As an example of the hardware implementation of the adaptive UAV intelligent navigation system provided in this embodiment of the invention, the device provided in this embodiment of the invention can be directly executed by a processor 301 in the form of a hardware decoding processor. For example, it can be executed by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components to implement the adaptive UAV intelligent navigation method provided in this embodiment of the invention.

[0146] The memory 302 in this embodiment of the invention is used to store various types of data to support the operation of the adaptive UAV intelligent navigation system, or to store data for execution. Figure 1The program code for the method shown. Examples of this data include: any executable instructions for operation on an adaptive UAV intelligent navigation system, such as executable instructions, and programs implementing the adaptive UAV intelligent navigation method of embodiments of the present invention may be contained in executable instructions.

[0147] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including functions for executing... Figure 1 The program code for the method shown. In such an embodiment, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by the central processing unit, it performs the various functions defined in the apparatus of this application.

[0148] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 Figure 1 A device that provides the functions specified in one or more boxes.

[0149] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive intelligent navigation method for unmanned aerial vehicles (UAVs), characterized in that, include: A basic navigation model is trained, and the perplexity of the generated navigation path points is calculated. Navigation path points with perplexity higher than a set threshold are marked as high perplexity path points. The perplexity is calculated based on a weighted average of action entropy and environmental semantic matching degree. The representation function of perplexity is: ； in, express The level of confusion at any given moment; and There are two hyperparameters; It is a moment The degree of semantic matching in the context of the environment. Indicates time The entropy of action; The environmental semantic matching degree is calculated by analyzing the observed images using a target detection model, and then calculating the average confidence score of targets related to the language instructions. The calculation formula is as follows: ； in, It is a moment The degree of semantic matching in the context of the environment. This indicates the target detection model's detection of objects within the current field of view that match the instructions. The number of relevant targets It is the first Confidence score of each detected target; An altitude adjustment prediction module is added to the basic navigation model, and the basic navigation model with the added altitude adjustment prediction module is retrained using marked high confusion waypoints; the altitude adjustment prediction module determines whether to trigger altitude adjustment based on whether the navigation waypoint is a high confusion waypoint. By utilizing a well-trained basic navigation model, altitude adjustments can be automatically triggered when decision-making is difficult, enabling the UAV to proactively change its flight altitude in complex environments in order to obtain observation information from the optimal perspective.

2. The adaptive UAV intelligent navigation method as described in claim 1, characterized in that, Action entropy is calculated from the action probability distribution output by the model. The higher the action entropy, the greater the uncertainty in the decision. The calculation formula is: ； in, Indicates time The entropy of action, Represents the set of all navigation actions. Indicates the model at time... Predict the execution action The probability of.

3. The adaptive UAV intelligent navigation method as described in claim 1, characterized in that, The total loss function of the base navigation model after adding the altitude adjustment prediction module is set as follows: Its expression is: ； ； ； in, The loss function during navigation; For high-level decision loss function; The model predicts the actions of the expert. The probability of; Indicates the total number of path points. Representing path points In category The actual label below, if the path point An altitude adjustment should be triggered. ,otherwise ; For at the path point Category The height adjustment probability distribution of the forecast. This indicates the probability that the height does not need to be adjusted. This indicates the probability that the altitude needs to be adjusted.

4. The adaptive UAV intelligent navigation method as described in claim 1, characterized in that, In the process of automatically triggering altitude adjustments when decision-making is difficult using a trained basic navigation model, when the final target is detected in the UAV's field of view, in order to obtain better observation information, the UAV should move closer to the target altitude. The UAV's new altitude... Represented as: ； in, Represents the target object in the world coordinate system The axis coordinate value, i.e., the height value.

5. The adaptive unmanned aerial vehicle (UAV) intelligent navigation method as described in claim 1, characterized in that, In the process of automatically triggering altitude adjustments when decision-making is difficult using a pre-trained basic navigation model, if no target is detected, the drone will perform an ascent to expand the search range, and the drone will reach a new altitude. Based on adaptive determination according to environmental conditions: ； in, Indicates the drone's current altitude; Represents the set of other observed objects; Represents objects Height; , It is a constant; This represents the maximum value function.

6. An adaptive unmanned aerial vehicle (UAV) intelligent navigation system, characterized in that, include: The high-confusion pathpoint marking module is used to train the base navigation model. It calculates the confusion degree of the generated navigation pathpoints and marks navigation pathpoints with confusion degrees higher than a set threshold as high-confusion pathpoints. The confusion degree is calculated based on a weighted average of action entropy and environmental semantic matching degree. The representation function of perplexity is: ； in, express The level of confusion at any given moment; and There are two hyperparameters; It is a moment The degree of semantic matching in the context of the environment. Indicates time The entropy of action; The environmental semantic matching degree is calculated by analyzing the observed images using a target detection model, and then calculating the average confidence score of targets related to the language instructions. The calculation formula is as follows: ； in, It is a moment The degree of semantic matching in the context of the environment. This indicates the target detection model's detection of objects within the current field of view that match the instructions. The number of relevant targets It is the first Confidence score of each detected target; The basic navigation model retraining module is used to add an altitude adjustment prediction module to the basic navigation model. It retrains the basic navigation model after adding the altitude adjustment prediction module using marked high confusion waypoints. The altitude adjustment prediction module determines whether to trigger altitude adjustment based on whether the navigation waypoint is a high confusion waypoint. The adaptive altitude adjustment module is used to automatically trigger altitude adjustment when decision-making is difficult by utilizing a trained basic navigation model, enabling the UAV to actively change its flight altitude in complex environments in order to obtain observation information from the optimal perspective.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the adaptive unmanned aerial vehicle intelligent navigation method as described in any one of claims 1-5.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the adaptive unmanned aerial vehicle intelligent navigation method as described in any one of claims 1-5.

Images