Artificial intelligence-based multi-compatible unmanned aerial vehicle operating system

Abstract

This invention relates to the field of unmanned aerial vehicle (UAV) control technology, specifically providing an AI-based multi-compatible UAV operating system. The system includes a standardized external interface unit pre-embedded in the UAV body and a pluggable intelligent adapter module. The intelligent adapter module has a built-in edge computing chip and runs a heterogeneous instruction parsing unit and a dynamic mapping reconstruction unit. The heterogeneous instruction parsing unit collects UAV bus data streams from the intelligent adapter module, constructs an initial environmental feature matrix, and generates protocol identification tags using a protocol recognition neural network model. The dynamic mapping reconstruction unit calls the protocol conversion master program based on the protocol identification tags, mapping standard control commands to target control commands with a proprietary protocol format. The flight status feedback module performs dynamic calibration using an adaptive PID compensation network. This invention requires no modification to the original firmware, enabling plug-and-play control of multiple brands of UAVs using a single external control system.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, specifically to a multi-compatible UAV operating system based on artificial intelligence. Background Technology

[0002] With the development of drone technology, its application in industries such as power line inspection, geographic surveying, and logistics delivery is becoming increasingly widespread. Currently, drone products on the market are manufactured by different companies, each using closed proprietary communication protocols and flight control logic. This protocol heterogeneity leads to significant compatibility barriers between different brands of drones; the ground control terminal of one manufacturer cannot directly establish an effective control link with a drone from another manufacturer. For industry users with multi-brand mixed fleets, dedicated control equipment must be provided for each model, increasing equipment procurement costs and reducing operational portability and efficiency.

[0003] Existing cross-platform control technologies primarily focus on firmware modification at the software level or protocol conversion based on external computing devices. Software cracking solutions require modifying the internal firmware code of the drone's flight controller, compromising the integrity of the original system. This not only poses a high risk of flight loss of control but also voids the equipment warranty. While protocol conversion solutions based on external devices retain the original firmware, they typically rely on a portable computer as an intermediary for command decoding and repackaging. Limited by the computing architecture of the intermediate device, this conversion process increases signal transmission latency. Under highly dynamic flight conditions, this latency severely reduces the system's response speed and control stability. Existing solutions mostly employ static command mapping, which is difficult to adapt to the differences in physical characteristics between different drone models and cannot meet the stringent requirements of real-time performance and consistency for cross-platform control.

[0004] Therefore, there is a lack of a multi-compatible drone operating system in the existing technology that is based on hardware modularity, low latency, and adaptive capabilities. Summary of the Invention

[0005] This invention overcomes the shortcomings of existing technologies and provides a multi-compatible drone operating system based on artificial intelligence. By reserving standardized physical interfaces on the drone end and combining them with pluggable intelligent external modules, a complete technical chain of "physical access - protocol recognition - semantic conversion - closed-loop feedback" is constructed, realizing plug-and-play control of multiple brands of drones and intelligent flight feel transfer using a single external control system. This effectively solves the problems existing in existing technologies.

[0006] The technical solution adopted by this invention is as follows: This solution provides a multi-compatible UAV operating system based on artificial intelligence, which is applied between the UAV body and the external target control system, including a standardized external interface unit, an intelligent adaptation module, a heterogeneous instruction parsing unit, a dynamic mapping reconstruction unit, and a flight status feedback module.

[0007] The standardized external interface unit is pre-embedded and integrated into the fuselage structure or payload bay of the UAV, physically connecting to the data bus and power management circuitry of the original UAV flight controller. When the intelligent adapter module is inserted, the standardized external interface unit establishes a physical electrical connection between the intelligent adapter module and the UAV flight controller, including a power supply path and a bidirectional data transmission link, acquiring the UAV bus data stream. The standardized external interface unit internally integrates level conversion circuitry and hardware identification pins.

[0008] The intelligent adapter module adopts a pluggable hardware form factor and integrates an edge computing chip, a multimodal communication module, and a power management circuit. After being inserted into a standardized external interface unit, the intelligent adapter module sends a preset handshake signal through a hardware identification pin to activate the UAV flight control and establishes a connection with an external target control system through the multimodal communication module.

[0009] The heterogeneous instruction parsing unit runs on the edge computing chip of the intelligent adaptation module. By monitoring the UAV bus data stream, the heterogeneous instruction parsing unit collects the UAV flight controller's heartbeat packets, sensor data frames, and firmware version information during standby mode to construct an initial environmental feature matrix. The unit also incorporates a protocol recognition neural network model, which extracts and classifies features from the initial environmental feature matrix to identify the current UAV's brand, model, and the syntax structure of its proprietary communication protocol, generating protocol identification tags.

[0010] The dynamic mapping and reconstruction unit, based on the protocol identification tags output by the heterogeneous instruction parsing unit, calls the corresponding protocol conversion master program from a local or cloud database. The dynamic mapping and reconstruction unit includes an instruction semantic alignment submodule, a parameter compensation submodule, an instruction encapsulation submodule, and an adaptive PID compensation network. When the intelligent adaptation module receives standard control commands (such as throttle, roll, pitch, and yaw) from an external target control system, the dynamic mapping and reconstruction unit inputs the standard control commands into a pre-built instruction semantic space conversion model. Combining this with the UAV's real-time flight status, it calculates a target control command sequence conforming to the current UAV's proprietary protocol format and sends it to the UAV flight controller via a standardized external interface unit.

[0011] The flight status feedback module acquires the actual attitude feedback data and sensor data of the UAV after executing target control commands in real time through a standardized external interface unit. It uses a Kalman filter algorithm to denoise and align the actual attitude feedback data and sensor data in terms of time sequence to generate a status feedback vector. The status feedback vector is then input into the adaptive PID compensation network in the dynamic mapping reconstruction unit. The adaptive PID compensation network automatically adjusts the mapping parameters of the command semantic space transformation model according to the deviation between the status feedback vector and the expected commands of the external target control system, thereby achieving dynamic calibration and collaborative optimization of the control feel.

[0012] Furthermore, the power management circuit of the intelligent adapter module converts the high voltage of the drone battery into the standard operating voltage required by the intelligent adapter module, and has overcurrent protection and reverse connection protection functions; the level conversion circuit of the standardized external interface unit converts between the CMOS level of the drone flight controller and the RS-232 or TTL level of the intelligent adapter module to ensure the integrity and stability of the data signal during transmission; the hardware identification pin is connected to the general-purpose I / O port of the drone flight controller through a voltage divider circuit. When the intelligent adapter module is inserted, the change in physical level triggers the drone flight controller to interrupt, so that the drone flight controller automatically switches from normal flight mode to external compatibility mode, opening the read and write permissions of the underlying bus.

[0013] Furthermore, the workflow of the heterogeneous instruction parsing unit specifically includes the following steps:

[0014] Step S1, Data Acquisition: After the intelligent adapter module is inserted, it collects the data stream of the UAV bus within a preset time window through the standardized external interface unit to obtain the raw byte stream data. The raw byte stream data contains the frame header, frame tail, check bit and function code distribution pattern of the data frame.

[0015] Step S2, Frame Structure Analysis: The original byte stream data is segmented using the sliding window algorithm to identify the distribution patterns of the frame header, frame tail, checksum, and function code, and to extract message timing features. Simultaneously, firmware version information, heartbeat packet features, and sensor data features are parsed from the original byte stream data. The message timing features, firmware version information, heartbeat packet features, and sensor data features are then fused using multi-dimensional features to construct an initial environmental feature matrix.

[0016] Step S3, Protocol Fingerprint Matching: Input the initial environmental feature matrix into the protocol recognition neural network model. The protocol recognition neural network model is trained with a large number of brand drone protocol samples and can output the brand probability distribution of the current drone, the syntax structure of the private communication protocol, and the protocol recognition label. Select the brand with the highest probability, the corresponding syntax structure, and the protocol recognition label as the final recognition result.

[0017] Step S4, Verification and Locking: The identified protocol identification tags and the syntax structure of the private communication protocol are compared and verified with the standard protocol features in the local database. If the matching degree is higher than the preset threshold, the protocol identification tags and the syntax structure of the private communication protocol are locked. If the matching degree is lower than the preset threshold, a query request is sent to the cloud to download the latest protocol configuration package and update the local database.

[0018] Furthermore, the dynamic mapping reconstruction unit comprises an instruction semantic alignment submodule, a parameter compensation submodule, and an instruction encapsulation submodule. The collaborative working process includes the following steps:

[0019] Step T1, Protocol Loading and Command Semantic Alignment: The dynamic mapping reconstruction unit identifies the label according to the protocol and calls the corresponding protocol conversion master program from the local or cloud database; the command semantic alignment submodule receives standard control commands from the external target control system, parses the standard control commands into a unified high-dimensional semantic vector, and inputs the high-dimensional semantic vector into the pre-built command semantic space conversion model;

[0020] Step T2, State Fusion and Parameter Mapping: The command semantic space transformation model, combined with the real-time flight state of the UAV, inputs the high-dimensional semantic vector into the parameter compensation submodule. The parameter compensation submodule, based on the characteristics of the current UAV model, calls a pre-stored dynamic parameter mapping table to map the control quantities in the high-dimensional semantic vector into execution parameters adapted to the physical characteristics of the target UAV. For example, it maps the standardized 0-100% throttle range to the PWM pulse width value or throttle percentage curve corresponding to a specific brand of UAV.

[0021] Step T3, Parameter Fine-tuning and Optimization: Combining the deviation between the state feedback vector input by the flight state feedback module and the expected command of the external target control system, the mapping parameters of the command semantic space transformation model are automatically adjusted using an adaptive PID compensation network to achieve fine-tuning of the execution parameters and generate calibrated optimized execution parameters.

[0022] Step T4, Instruction Encapsulation and Transmission: The instruction encapsulation submodule identifies the private protocol format corresponding to the tag according to the protocol, serializes the optimized execution parameters into a binary data stream, and adds a checksum and frame header to generate the final target control instruction sequence.

[0023] Furthermore, the flight status feedback module includes a status monitoring submodule, a data filtering submodule, and an adaptive compensation submodule. The status monitoring submodule reads real-time sensor data from the UAV via a standardized external interface unit, including three-axis gyroscope data, three-axis accelerometer data, GPS coordinate data, and barometer altitude data. The data filtering submodule uses a Kalman filter algorithm to fuse the sensor data, eliminating sensor noise and short-term interference, and calculates the UAV's current high-precision attitude angle, angular velocity, linear velocity, and position information to construct a status feedback vector. The adaptive compensation submodule receives the status feedback vector and inputs it into the adaptive PID compensation network in the dynamic mapping reconstruction unit. The adaptive PID compensation network automatically adjusts the mapping parameters of the command semantic space transformation model based on the deviation between the status feedback vector and the expected commands from the external target control system, achieving dynamic calibration and collaborative optimization of the control feel.

[0024] Furthermore, the system also includes a cloud-based co-evolution module. The intelligent adaptation module connects to the cloud-based co-evolution module via a multimodal communication module. The cloud-based co-evolution module collects unknown protocol feature data encountered by the heterogeneous command parsing unit during the identification process, uploaded by the intelligent adaptation module, as well as flight status and command response pairs recorded by the dynamic mapping reconstruction unit during the control process. The cloud-based co-evolution module utilizes deep reinforcement learning algorithms to train the collected data offline, continuously optimizing the accuracy of the protocol recognition neural network model and the adaptability of the command semantic space transformation model. The newly generated model parameters are sent to the intelligent adaptation module through an encrypted channel, enabling self-iteration and upgrades at the system software level, ensuring compatibility with newly released drone brands or firmware versions on the market.

[0025] Compared with the prior art, the beneficial effects of the present invention are:

[0026] (1) Through the hardware architecture of “standardized external interface unit + intelligent adaptation module”, the compatibility problem of drones is innovatively decoupled from complex software flashing to simple modular physical expansion, achieving compatibility without disassembling the drone or modifying the original firmware.

[0027] (2) By introducing a heterogeneous instruction parsing unit based on deep neural networks, the protocol recognition neural network model is used to extract features and classify the UAV bus data stream, thereby realizing the automated and intelligent recognition of private protocols. This breaks away from the inefficient mode of manual protocol cracking and enables the system to have strong generalization and adaptability, and can quickly adapt to different brands and models of UAVs.

[0028] (3) By constructing a closed-loop system of “dynamic mapping reconstruction-flight status feedback”, AI algorithms and adaptive PID compensation networks are used to solve the differences in dynamic characteristics between different aircraft models, realize intelligent alignment of cross-platform control feel, improve the accuracy, stability and consistency of flight control, meet the stringent requirements of cross-platform control for real-time performance and consistency, and avoid transmission delay problems caused by external equipment transfer.

[0029] (4) Through the cloud-based collaborative evolution mechanism, the collected data is trained offline using deep reinforcement learning algorithms and the generated new model parameters are sent to the intelligent adaptation module to realize the self-iteration and upgrade of the system software, giving the operating system the ability to continuously evolve and enabling it to be compatible with newly released drone brands or firmware versions on the market. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall architecture of a multi-compatible drone operating system based on artificial intelligence proposed in this invention;

[0031] Figure 2 This is a schematic diagram of the collaborative control logic between the dynamic mapping reconstruction unit and the flight status feedback module of the present invention;

[0032] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0033] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0034] Example 1:

[0035] Please see Figures 1-2 This embodiment presents an AI-based multi-compatible UAV operating system, applied to connect heterogeneous external target control systems with UAVs of different brands, enabling plug-and-play control across systems. It includes a standardized external interface unit, an intelligent adaptation module, a heterogeneous command parsing unit, a dynamic mapping reconstruction unit, a flight status feedback module, and a cloud-based collaborative evolution module.

[0036] The standardized external interface unit is physically integrated into a reserved compartment on the side of the drone's fuselage. This unit employs an industrial-grade aviation plug structure, including positive and negative power pins, two pairs of differential data transmission pins, and a hardware identification pin. The positive and negative power pins directly connect to the drone's battery power management circuitry and include an overcurrent protection fuse and voltage regulator circuitry within the interface. The differential data transmission pins connect to the original drone flight controller's debugging interface via a level conversion chip. The hardware identification pin connects to the drone flight controller's general-purpose I / O port via a voltage divider circuit. When the intelligent adapter module is inserted, the drone flight controller detects the level change on the hardware identification pin, triggering it to enter "external compatibility mode." In this mode, the drone flight controller gains access to the underlying data bus and allows the reception of externally injected control commands.

[0037] The intelligent adaptation module is the core processing unit of the system. Its shell shape matches the standardized external interface unit. Internally, the intelligent adaptation module integrates a high-performance edge computing chip, a large-capacity Flash memory, a multimodal communication module (supporting Wi-Fi, Bluetooth, and custom RF protocols), and a power management circuit. The edge computing chip, as the main control core, runs the algorithm logic of the heterogeneous instruction parsing unit and the dynamic mapping reconstruction unit. The Flash memory stores the protocol driver libraries and neural network model weight files for various brands of drones. The multimodal communication module establishes low-latency data links with external target control systems (tablets, dedicated remote controllers) and the cloud-based collaborative evolution module, receiving control commands, transmitting status information, and uploading unknown protocol data. The power management circuit converts the power voltage input from the drone into the operating voltage required by the various components within the intelligent adaptation module and monitors the voltage.

[0038] The heterogeneous instruction parsing unit is a software logic module running on the edge computing chip. After the intelligent adaptation module powers on, the heterogeneous instruction parsing unit performs "identity recognition" on the connected drone through a standardized external interface unit. Specifically, the heterogeneous instruction parsing unit configures the differential data transmission pin to a high-impedance input mode, passively listening to the data stream transmitted on the drone's bus. Within a preset listening time window (5 seconds), the heterogeneous instruction parsing unit collects raw byte stream data containing hundreds of data frames. The built-in message parser preprocesses the raw byte stream data, removing noise and empty frames, and filtering out valid control instruction frames and sensor data frames. It extracts the time-domain features (frame interval time, byte periodicity) and frequency-domain features (byte value entropy features) of the valid control instruction frames and sensor data frames, and parses firmware version information, heartbeat packet features, and sensor data features from the raw byte stream data to construct an initial environment feature matrix. This initial environment feature matrix is ​​input into the built-in protocol recognition neural network model, which outputs the brand and model of the current drone, the protocol version number, the syntax structure of the private communication protocol, and the protocol identification tag. The heterogeneous instruction parsing unit sends the protocol identification tag and the syntax structure of the private communication protocol to the dynamic mapping reconstruction unit as the basis for subsequent instruction conversion.

[0039] The dynamic mapping and reconstruction unit, also a core algorithm module running on the edge computing chip, implements the format conversion and logical mapping of control commands. The dynamic mapping and reconstruction unit includes a command semantic alignment submodule, a parameter compensation submodule, a command encapsulation submodule, and an adaptive PID compensation network. After receiving the protocol identification tag sent by the heterogeneous command parsing unit, the dynamic mapping and reconstruction unit retrieves the protocol conversion master control program corresponding to the protocol identification tag from a local or cloud database. The protocol conversion master control program includes a command semantic space conversion model and a dynamic parameter mapping table. The dynamic parameter mapping table stores the current UAV's physical limit parameters (maximum tilt angle, maximum climb rate) and control response characteristic curves. When the intelligent adaptation module receives standard control commands from an external target control system through the multimodal communication module, the dynamic mapping and reconstruction unit parses the standard control commands and, combined with the UAV's real-time flight status, calculates a target control command sequence conforming to the current UAV's private protocol format, and sends it to the UAV flight controller through a standardized external interface unit.

[0040] The flight status feedback module, comprising a status monitoring submodule, a data filtering submodule, and an adaptive compensation submodule, ensures that the actual response of the UAV is consistent with the command intent of the external target control system. The status monitoring submodule reads real-time sensor data from the UAV via a standardized external interface unit, including three-axis gyroscope data, three-axis accelerometer data, GPS coordinate data, and barometer altitude data. Upon receiving the sensor data, the data filtering submodule uses a Kalman filter algorithm to fuse the data, eliminating sensor noise and short-term interference, and calculates the UAV's current high-precision attitude angles (pitch, roll, yaw), angular velocity, linear velocity, and position information, constructing a status feedback vector. The adaptive compensation submodule receives the status feedback vector and inputs it into the adaptive PID compensation network in the dynamic mapping reconstruction unit. The adaptive PID compensation network automatically adjusts the mapping parameters of the command semantic space transformation model based on the deviation between the status feedback vector and the expected commands of the external target control system, achieving dynamic calibration and collaborative optimization of the control feel.

[0041] The cloud-based co-evolution module connects to the intelligent adaptation module via a multimodal communication module. This module addresses the compatibility issues caused by the continuous emergence of new drones. When the heterogeneous command parsing unit fails to recognize the drone protocol locally (i.e., confidence determination fails), the intelligent adaptation module automatically activates its learning mode. In learning mode, the intelligent adaptation module records the bus data stream when the drone receives a specific manual stimulus signal (user manually controls drone takeoff), along with the feature vectors generated during protocol recognition, and uploads this data to the cloud-based co-evolution module via the multimodal communication module. The cloud-based co-evolution module deploys a powerful deep learning cluster that performs offline analysis on the collected bus data stream and the unknown protocol feature data encountered by the heterogeneous command parsing unit during recognition. It manually annotates the protocol frame structure and trains new neural network model weights using deep reinforcement learning algorithms. After the update is complete, an incremental update package is generated, containing the new protocol driver library and the updated neural network model weights. The cloud-based co-evolution module then pushes this incremental update package to the intelligent adaptation module. The intelligent adaptation module receives incremental update packets, verifies the signature, and then updates the protocol library and model file in the local Flash memory, thus realizing the self-evolution of system functions.

[0042] Example 2:

[0043] This embodiment, based on Embodiment 1, details the specific implementation process of protocol identification by the heterogeneous instruction parsing unit. The heterogeneous instruction parsing unit immediately initiates the "protocol fingerprint acquisition" process the moment the intelligent adaptation module is inserted into the drone. Specific steps include:

[0044] Step S1, Data Capture:

[0045] After the intelligent adapter module is inserted, it collects the data stream of the UAV bus within a preset time window through a standardized external interface unit to obtain the raw byte stream data.

[0046] Step S2, Frame Structure Analysis:

[0047] The original byte stream data is segmented using a sliding window algorithm to identify the distribution patterns of frame headers, frame trailers, check bits, and function codes, and to extract message timing features. Simultaneously, firmware version information, heartbeat packet features, and sensor data features are parsed from the original byte stream data. The message timing features, firmware version information, heartbeat packet features, and sensor data features are then fused using multi-dimensional features to construct an initial environmental feature matrix.

[0048] Specifically, for each frame of data, the following features are extracted:

[0049] (1) Frame length distribution: Statistical data show the variation pattern of frame length. Different brands of drones have fixed frame length characteristics;

[0050] (2) Byte frequency distribution: Calculate the frequency of each byte value (0x00-0xFF) in the original byte stream data;

[0051] (3) Timing interval characteristics: The time interval between two consecutive data frames is measured;

[0052] (4) Check code characteristics: Identify the check algorithm type at the end of the frame;

[0053] Combine all features into a high-dimensional feature vector ;

[0054] Step S3, Protocol fingerprint matching:

[0055] The initial environmental feature matrix is ​​input into the protocol recognition neural network model. The protocol recognition neural network model is trained on a large number of brand drone protocol samples and can output the brand probability distribution, the syntax structure of the private communication protocol, and the protocol recognition label for the current drone. The brand with the highest probability, its corresponding syntax structure, and the protocol recognition label are selected as the final recognition result.

[0056] Specifically, the protocol recognition neural network model outputs a probability set belonging to each category. Heterogeneous instruction parsing unit selection probability set The maximum value in and will Compare with the preset confidence threshold (0.95). If If the confidence level is greater than or equal to the preset confidence threshold, then a judgment is made. The corresponding brand category is the actual brand of the current drone, and a specific protocol identification tag and the syntax structure of the private communication protocol are generated. If If the confidence level is less than the preset threshold, the device is determined to be an unknown model. The heterogeneous instruction parsing unit will then send a request to the cloud-based collaborative evolution module via the multimodal communication module to upload the feature vector. The system prompts the user to manually select the device model.

[0057] Step S4, Verification and Locking:

[0058] The identified protocol identification tags and the syntax structure of the private communication protocol are compared and verified with the standard protocol features in the local database. If the matching degree is higher than the preset threshold, the protocol configuration is locked; if the matching degree is lower than the preset threshold, a query request is sent to the cloud to download the latest protocol configuration package and update the local database.

[0059] Through the above steps, the heterogeneous instruction parsing unit can quickly and accurately identify the UAV's identity and its protocol syntax structure, providing accurate parameter indexes for the dynamic mapping reconstruction unit.

[0060] Example 3:

[0061] This embodiment, based on Embodiment 1, describes in detail the collaborative working logic between the dynamic mapping reconstruction unit and the flight status feedback module. Specific steps include:

[0062] Step T1: Protocol loading and instruction semantic alignment:

[0063] The dynamic mapping reconstruction unit identifies the label according to the protocol and calls the corresponding protocol conversion master program from the local or cloud database; the instruction semantic alignment submodule receives standard control instructions from the external target control system, parses the standard control instructions into a unified high-dimensional semantic vector, and inputs the high-dimensional semantic vector into the pre-built instruction semantic space conversion model.

[0064] Specifically, when the operator pushes a lever on the external target control system, the multimodal communication module of the intelligent adaptation module receives the radio signal, demodulates it to obtain the raw control command packet, and then the dynamic mapping reconstruction unit parses the raw control command packet and transforms it into a standardized control vector defined within the system. ,in Representative control quantity (range normalized to between -1 and 1);

[0065] Step T2, State Fusion and Parameter Mapping:

[0066] The command semantic space transformation model combines the real-time flight status of the UAV and inputs the high-dimensional semantic vector into the parameter compensation submodule. The parameter compensation submodule calls the pre-stored dynamic parameter mapping table according to the current UAV model characteristics and maps the control quantities in the high-dimensional semantic vector into execution parameters that are adapted to the physical characteristics of the target UAV.

[0067] Considering that the flight control systems of different brands of drones respond differently to the same input signal, a mapping function is constructed to eliminate this difference. :

[0068] ;

[0069] in, This is the mapping parameter matrix learned for the current drone model. Mapping function. The flight feel desired by the external target control system was simulated by standardizing the control vectors. Mapped to intermediate control variables suitable for the current drone Mapping function It can be achieved through neural network fitting, or through a combination of table lookup and linear interpolation;

[0070] In this embodiment, if the throttle command issued by the external target control system is 50%, and the current throttle response curve of the UAV is non-linear, the dynamic mapping reconstruction unit converts it into the corresponding PWM pulse width value according to the dynamic parameter mapping table.

[0071] Step T3, Parameter Fine-tuning and Optimization:

[0072] By combining the deviation between the state feedback vector input from the flight state feedback module and the expected commands from the external target control system, the mapping parameters of the command semantic space transformation model are automatically adjusted using an adaptive PID compensation network to achieve fine-tuning of the execution parameters and generate corrected optimized execution parameters.

[0073] In this embodiment, the data filtering submodule of the flight state feedback module outputs the current state feedback vector of the UAV. ,in Represents speed, , , These represent the roll angle, pitch angle, and yaw angle, respectively. The adaptive compensation submodule calculates the state feedback vector. Error vector between the previous time step and the desired state of the instruction For example, the desired pitch angle is... The actual pitch angle is Then pitch error To eliminate this error, the adaptive compensation submodule runs a corrected model:

[0074] ;

[0075] in, It is a parameter-adjustable proportional-integral-derivative controller used to calculate the correction amount. Correction amount This indicates the additional control input required to eliminate the current flight error;

[0076] Step T4, Instruction Encapsulation and Transmission:

[0077] The dynamic mapping reconfiguration unit will transfer intermediate control quantities With correction amount Perform weighted fusion to generate target control commands. :

[0078] ;

[0079] in, This is a weighting coefficient that is dynamically adjusted based on the flight mode. In precision hovering mode, The value is relatively small, and feedback compensation dominates to enhance stability; in high-speed maneuvering mode, The value is relatively large, and it mainly responds to the operator's commands;

[0080] Finally, the dynamic mapping reconstruction unit, based on the private protocol format corresponding to the protocol identification tag, transmits the target control commands. The data is serialized into a binary data stream, and a checksum and frame header are added to generate the final target control command sequence, which is then sent to the UAV flight controller through a standardized external interface unit.

[0081] Step T5, closed-loop iteration:

[0082] When the drone performs Subsequently, its flight status changed, the status monitoring submodule captured the new status data, and regenerated the status feedback vector. The system then enters the next control cycle. Through this feedback mechanism, the intelligent adaptation module not only achieves compatibility in command formats but also intelligent adaptation of control performance, ensuring that operators can obtain a highly consistent and accurate flight experience when using the same external target control system to operate drones of different brands.

[0083] Example 4:

[0084] This embodiment illustrates the effects of the present invention in conjunction with a specific application scenario.

[0085] Suppose a power inspection company owns quadcopter drones of brand A and fixed-wing drones of brand B. The inspectors want to use the uniformly provided ground station software and remote controller of brand C for their work.

[0086] The inspector first inserts the smart adapter module, pre-loaded with the Brand A driving protocol, into the standardized external interface unit of the Brand A drone. Upon powering on the smart adapter module, the heterogeneous command parsing unit identifies it as a Brand A model within 2 seconds. The inspector then opens the Brand C ground station and clicks "Connect," establishing a link between the smart adapter module and the ground station. The dynamic mapping and reconstructing unit begins receiving commands from Brand C and translating them into Brand A's proprietary commands. When the inspector pushes the throttle for takeoff, the flight status feedback module monitors the Brand A's ascent speed in real time and, through adaptive compensation, eliminates the insufficient lift issue caused by the aging of Brand A's motors, ensuring a smooth takeoff for the drone.

[0087] During operations, the inspector needs to switch to a brand B fixed-wing drone for long-distance patrols. The inspector simply needs to remove the smart adapter module from the brand A drone, replace it with a smart adapter module optimized for brand B, and plug it into the brand B drone's interface. The system automatically recognizes brand B and loads the corresponding dynamic parameters. The inspector doesn't need to change their remote controller and can still use the familiar brand C operating logic. The dynamic mapping and reconfiguration unit automatically handles the significant differences in flight control principles between quadcopters and fixed-wing aircraft (such as the fixed-wing aircraft using throttle to control speed rather than altitude), allowing the inspector to seamlessly switch equipment for operations.

[0088] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A multi-compatible unmanned aerial vehicle (UAV) operating system based on artificial intelligence, characterized in that, This includes a standardized external interface unit and an intelligent adapter module; the standardized external interface unit is pre-embedded and integrated into the UAV body, connecting to the UAV flight control data bus to acquire the UAV bus data stream; The standardized external interface unit integrates hardware identification pins; the intelligent adapter module has a built-in edge computing chip and a multimodal communication module. After the intelligent adaptation module is inserted into the standardized external interface unit, it activates the UAV flight control through hardware identification pins and connects to the external target control system through the multimodal communication module. The edge computing chip runs a heterogeneous instruction parsing unit and a dynamic mapping reconstruction unit. The heterogeneous instruction parsing unit collects the UAV bus data stream, constructs an initial environmental feature matrix, and uses the built-in protocol recognition neural network model to extract and classify features from the initial environmental feature matrix, generating protocol recognition tags. The dynamic mapping reconstruction unit calls the protocol conversion master program according to the protocol recognition tags. When the intelligent adaptation module receives standard control commands from the external target control system, the dynamic mapping reconstruction unit calculates the target control command sequence based on the UAV's real-time flight status and sends it to the UAV flight control through the standardized external interface unit.

2. The multi-compatible unmanned aerial vehicle operating system based on artificial intelligence according to claim 1, characterized in that: The specific working steps of the heterogeneous instruction parsing unit include: Step S1, Data Acquisition: After the intelligent adapter module is inserted, it acquires the UAV bus data stream through the standardized external interface unit to obtain the raw byte stream data; the raw byte stream data contains the frame header, frame tail, check bit, and function code distribution pattern of the data frame; Step S2, Frame Structure Analysis: The original byte stream data is segmented, and the distribution patterns of the frame header, frame tail, check bit, and function code are identified to extract message timing features. At the same time, firmware version information, heartbeat packet features, and sensor data features are parsed from the original byte stream data. The message timing features, firmware version information, heartbeat packet features, and sensor data features are fused into multi-dimensional features to construct an initial environment feature matrix. Step S3, Protocol Fingerprint Matching: Input the initial environmental feature matrix into the protocol recognition neural network model. The protocol recognition neural network model outputs the brand probability distribution of the current drone, the syntax structure of the private communication protocol, and the protocol recognition label. Select the drone brand, the syntax structure of the private communication protocol, and the protocol recognition label corresponding to the highest probability as the final recognition result. Step S4, Verification and Locking: The final identification result is compared and verified with the standard protocol features in the local database. If the matching degree is higher than the preset threshold, the protocol identification tag and the syntax structure of the private communication protocol are locked. If the matching degree is lower than the preset threshold, a query request is sent to the cloud to download the latest protocol configuration package and update the local database.

3. The multi-compatible unmanned aerial vehicle operating system based on artificial intelligence according to claim 2, characterized in that: The multi-compatible UAV operating system also includes a flight status feedback module. The flight status feedback module acquires the actual attitude feedback data and sensor data of the UAV after executing the target control command in real time through a standardized external interface unit, and generates a status feedback vector using a Kalman filter algorithm.

4. The multi-compatible unmanned aerial vehicle operating system based on artificial intelligence according to claim 3, characterized in that: The dynamic mapping reconstruction unit includes an instruction semantic alignment submodule, a parameter compensation submodule, an instruction encapsulation submodule, and an adaptive PID compensation network. The instruction semantic alignment submodule receives standard control instructions from an external target control system, parses the standard control instructions into high-dimensional semantic vectors, and inputs the high-dimensional semantic vectors into a pre-built instruction semantic space transformation model. The parameter compensation submodule calls a pre-stored dynamic parameter mapping table to map the control quantities in the high-dimensional semantic vectors into execution parameters adapted to the physical characteristics of the UAV. The adaptive PID compensation network automatically adjusts the mapping parameters of the instruction semantic space transformation model based on the deviation between the state feedback vector and the expected instructions of the external target control system, generating corrected optimized execution parameters. The instruction encapsulation submodule identifies the private protocol format corresponding to the tag according to the protocol, serializes the corrected and optimized execution parameters into a binary data stream, and adds a check code and frame header to generate the final target control instruction sequence.

5. The multi-compatible unmanned aerial vehicle operating system based on artificial intelligence according to claim 4, characterized in that: The multi-compatible drone operating system also includes a cloud-based collaborative evolution module; The cloud-based co-evolution module connects to the intelligent adaptation module through a multimodal communication module. The cloud-based co-evolution module collects unknown protocol feature data encountered by the heterogeneous instruction parsing unit during the identification process, as well as flight status and instruction response pair data recorded by the dynamic mapping reconstruction unit during the control process. The cloud-based co-evolution module performs offline training on the collected unknown protocol feature data, flight status and instruction response pair data, generates new model parameters, and sends them to the intelligent adaptation module.

6. The multi-compatible unmanned aerial vehicle operating system based on artificial intelligence according to claim 4, characterized in that: The sensor data acquired by the flight status feedback module includes three-axis gyroscope data, three-axis accelerometer data, GPS coordinate data, and barometer altitude data. The Kalman filter algorithm fuses the sensor data to calculate high-precision attitude angles, angular velocities, linear velocities, and position information, and constructs a status feedback vector.

Images