Adaptive flight control method and related equipment based on dynamic alternation.

By constructing a multi-dimensional state space and generating a set of multi-domain alternating strategies, the perception, communication, and computing modes of the UAV are dynamically and alternately controlled, solving the problem of balancing and optimizing resources and performance of the UAV in complex low-altitude environments, and improving environmental adaptability and operational stability.

CN123113371APending Publication Date: 2026-11-17PING AN TECH (SHENZHEN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611448369.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-17
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing UAV control systems struggle to achieve a balance between performance and resources when facing complex and ever-changing low-altitude operating environments, resulting in low environmental adaptability and operational stability.

Method used

By continuously acquiring multi-source state data to construct a multi-dimensional state space, and using reinforcement learning to generate a set of multi-domain alternating strategies, the perception, communication, and computing modes of the UAV are dynamically and collaboratively controlled to make decisions and control with the goal of optimizing mission efficiency.

Benefits of technology

It achieves adaptive and collaborative scheduling of multi-domain resources, improves the environmental adaptability and operational stability of UAV flight control, and adapts to complex dynamic low-altitude environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123113371A_ABST
    Figure CN123113371A_ABST
Patent Text Reader

Abstract

This invention relates to the field of artificial intelligence and discloses an adaptive flight control method and related equipment based on dynamic alternating orchestration. The method includes: continuously collecting multi-source state data under operational scenarios and standardizing it into multi-dimensional state representations to construct a multi-dimensional state space; inputting the state representation of the current time slot into a pre-trained reinforcement learning policy network to generate a multi-domain alternating policy set for the next time slot with the goal of optimizing task performance; performing coordinated dynamic alternating control on the perception operation mode, communication transmission mode, and computation offloading mode according to the policy set, completing environmental perception and inference calculations, and outputting the calculation results; controlling the UAV to perform corresponding actions based on the calculation results, and synchronously updating the state space of the next time slot. This invention can be applied to business scenarios such as fintech and healthcare, realizing adaptive coordinated scheduling of multi-domain resources, effectively adapting to complex dynamic low-altitude environments, and improving the environmental adaptability and operational stability of UAV flight control.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology and can be applied to business areas such as fintech and healthcare, particularly to an adaptive flight control method and related equipment based on dynamic alternating arrangement. Background Technology

[0002] With the rapid development of the low-altitude economy, drones, with their advantages of high mobility and flexible deployment, have gradually expanded their applications from traditional aerial photography and industrial inspection to multiple fields such as financial services and healthcare. For example, in the financial sector, drones can support business scenarios such as property insurance disaster site investigation, security inspection of financial outlets and important infrastructure, and can quickly acquire on-site images and spatial data, significantly improving the efficiency of risk verification and the speed of emergency response. In the healthcare sector, drones can be used for the transportation of emergency medicines and medical supplies, as well as on-site environmental monitoring of public emergencies, providing support for emergency treatment and public health protection.

[0003] In various application scenarios of drones, the drone control system is required to have closed-loop collaborative capabilities of perception, communication, computing and control in order to cope with the complex and ever-changing low-altitude operating environment. That is, the drone obtains environmental information through onboard sensors, transmits the data back to the edge or cloud for processing through the communication link, and then generates control commands based on the calculation results and sends them to the drone to execute actions. This closed-loop interaction is the core operating mechanism of low-altitude intelligence.

[0004] However, existing UAV operation control systems mostly adopt fixed operation modes and resource allocation methods. Each link is optimized independently and lacks a coordinated control mechanism. When facing complex operation scenarios and sudden changes, it is difficult to achieve a balance between performance and resources at the overall level, which affects the environmental adaptability and operational stability of UAV flight control. Summary of the Invention

[0005] The main objective of this invention is to provide an adaptive flight control method and related equipment based on dynamic alternating arrangement, aiming to solve the technical problem in the prior art that it is difficult to adapt to the low-altitude flight environment and the dynamic and ever-changing airborne resources, resulting in low environmental adaptability and operational stability of UAV flight control.

[0006] The technical solution of the present invention is as follows: The first aspect of this invention provides an adaptive flight control method based on dynamic alternating arrangement, comprising: Continuously acquire multi-source state data of drones in operational scenarios and perform standardized processing; construct a corresponding multi-dimensional state space based on the processed multi-dimensional state representation. The multidimensional state representation observed in the current time slot in the multidimensional state space is input into the policy network obtained in advance through reinforcement learning for decision reasoning, and a multi-domain alternating policy set for the next time slot is generated with the goal of optimizing task performance. Based on the multi-domain alternation strategy set, the perception operation mode, communication transmission mode and computation offloading mode of the UAV are controlled in a coordinated and dynamic alternation, respectively, to perceive the environment and output the corresponding calculation results. Based on the calculation results, the UAV is controlled to perform corresponding actions in the next time slot, and new multidimensional state representations are obtained and processed as multidimensional state representations for the next time slot, and the multidimensional state space is updated.

[0007] A second aspect of the present invention provides an adaptive flight control device based on dynamic alternating arrangement, comprising: The state space construction module is used to continuously acquire multi-source state data of UAVs in the operation scenario and perform standardized processing, and construct the corresponding multi-dimensional state space based on the processed multi-dimensional state representation. The strategy decision module is used to input the multidimensional state representation observed in the current time slot in the multidimensional state space into the policy network obtained in advance through reinforcement learning for decision reasoning, and generate a multi-domain alternating policy set for the next time slot with the goal of optimizing task performance. The dynamic alternation orchestration module is used to perform coordinated dynamic alternation control of the UAV's perception operation mode, communication transmission mode and computing offload mode according to the multi-domain alternation strategy set, perceive the environment and output the corresponding calculation results. The control execution module is used to control the UAV to perform corresponding actions in the next time slot according to the calculation results, continue to acquire and process new multidimensional state representations as multidimensional state representations for the next time slot, and update the multidimensional state space.

[0008] A third aspect of the present invention provides a computer device including at least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to perform the aforementioned adaptive flight control method based on dynamic alternation.

[0009] A fourth aspect of the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by one or more processors, cause the one or more processors to perform the above-described adaptive flight control method based on dynamic alternation.

[0010] Beneficial Effects: This invention discloses an adaptive flight control method and related equipment based on dynamic alternating orchestration. Compared with existing technologies, this invention continuously collects multi-source state data under operational scenarios and standardizes it into multi-dimensional state representations to construct a multi-dimensional state space. The state representation of the current time slot is input into a pre-trained reinforcement learning policy network to generate a multi-domain alternating policy set for the next time slot with the goal of optimizing task performance. Based on the policy set, collaborative dynamic alternating control is performed on the perception operation mode, communication transmission mode, and computation offloading mode to complete environmental perception and inference calculations and output the calculation results. Based on the calculation results, the UAV is controlled to perform corresponding actions, synchronously updating the state space of the next time slot. This invention can be applied to business scenarios such as fintech and healthcare, realizing adaptive collaborative scheduling of multi-domain resources, effectively adapting to complex dynamic low-altitude environments, and improving the environmental adaptability and operational stability of UAV flight control. Attached Figure Description

[0011] To more clearly illustrate the solutions in this invention, the accompanying drawings used in the description of the embodiments of this invention will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0012] Figure 1 A schematic diagram of an application environment for the adaptive flight control method based on dynamic alternating arrangement provided in an embodiment of the present invention; Figure 2 A flowchart of an adaptive flight control method based on dynamic alternating arrangement provided in an embodiment of the present invention; Figure 3 This is a flowchart of step S201 in the adaptive flight control method based on dynamic alternating arrangement provided in an embodiment of the present invention; Figure 4 This is a flowchart of step S203 in the adaptive flight control method based on dynamic alternating arrangement provided in an embodiment of the present invention; Figure 5 This is a flowchart of step S401 in the adaptive flight control method based on dynamic alternating arrangement provided in an embodiment of the present invention; Figure 6 This is a flowchart of step S402 in the adaptive flight control method based on dynamic alternating arrangement provided in an embodiment of the present invention; Figure 7 This is a flowchart of step S403 in the adaptive flight control method based on dynamic alternating arrangement provided in an embodiment of the present invention; Figure 8This is a flowchart of step S204 in the adaptive flight control method based on dynamic alternating arrangement provided in an embodiment of the present invention. Figure 9 A schematic diagram of the functional modules of an adaptive flight control device based on dynamic alternating arrangement provided in an embodiment of the present invention; Figure 10 A schematic diagram of the hardware structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention is further described in detail below. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. The embodiments of the invention are described below in conjunction with the accompanying drawings.

[0014] The adaptive flight control method based on dynamic alternating arrangement provided in this invention can be applied to, for example... Figure 1 In this application environment, it includes a drone swarm 101, a ground base station 102, an edge computing server 103, a cloud management platform 104, and a task management system 105. The ground base station 102 provides the medium for air-to-ground communication between the drone swarm 101 and the edge computing server 103, supporting uplink transmission of sensing data and downlink distribution of control commands. The edge computing server 103 is responsible for processing sensing data locally, performing real-time inference, and alternating strategy decisions. The cloud management platform 104 undertakes global fine-grained reconstruction, long-term path planning, and large-scale computing power scheduling tasks. The task management system 105 is used to issue job tasks and priority parameters.

[0015] Users can issue work instructions through the task management system 105, and the drone swarm 101 will be dispatched by the cloud management platform 104 and the edge computing server 103 to perform corresponding flight operations. These operations cover a variety of scenarios, including disaster insurance surveys, collateral asset inspections, and financial facility security in the financial field, as well as emergency drug transport, biological sample delivery, and on-site perception of public emergencies in the medical and health field.

[0016] The drone swarm 101 can be a multi-rotor drone, fixed-wing drone, or compound-wing drone with airborne sensing, computing, and communication capabilities, equipped with various sensing sensors such as synthetic aperture radar and optical cameras, as well as an airborne processing unit. The edge computing server 103 can be a local computing node deployed on the ground base station side, providing low-latency computing services; the cloud management platform 104 can be a cloud server, providing large-scale computing power and global data storage capabilities, or it can be a server for a distributed system.

[0017] It should be understood that the numbers of drones, base stations, edge servers, and cloud computing nodes mentioned above are merely illustrative. Depending on implementation needs, any number of operational nodes and computing nodes can be included.

[0018] like Figure 2 As shown, the adaptive flight control method based on dynamic alternating arrangement provided in this embodiment of the invention specifically includes the following steps: S201. Continuously acquire multi-source state data of the UAV in the operation scenario and perform standardized processing, and construct the corresponding multi-dimensional state space based on the processed multi-dimensional state representation.

[0019] In this embodiment, multi-source heterogeneous status data covering the entire operation link is continuously collected at fixed time slots. That is, a fixed-duration time slot is used as the basic scheduling cycle, for example, 100ms as a decision time slot. At the beginning of each time slot, a full set of status data is collected. The specific sources of the multi-source status data include multiple entities such as UAV onboard units, ground communication modules, computing nodes, and task systems.

[0020] After acquiring multi-source heterogeneous state data, normalization and encoding are performed on the original state data with different dimensions and data formats to generate a multi-dimensional state representation vector with a unified format and fixed dimensions. The multi-dimensional state representation vector is used as the basic state unit and combined with time-series accumulated data to construct a multi-dimensional state space. This results in a set of states composed of multi-dimensional state representations and time-series correlations in continuous time slots, which serves as the state basis for subsequent reinforcement learning-based decision reasoning.

[0021] For example, in the scenario of flood disaster insurance claims in the financial sector, the surrounding environment perception data of the claims drone, the quality of air-to-ground link communication, the load status of edge claims computing nodes, and the urgency level of the claims task can be continuously acquired. After standardization processing, a multi-dimensional state space can be constructed to comprehensively reflect the full-dimensional operation status of the current claims operation.

[0022] In emergency drug transport scenarios in the healthcare field, the system continuously acquires airspace environment data, air-to-ground communication channel status, operational load of edge medical scheduling computing nodes, and transport task priorities of transport drones. After processing, a state space is constructed to support adaptive scheduling of the transport process.

[0023] This embodiment standardizes and integrates multi-source heterogeneous state data in UAV operation scenarios to construct a unified multi-dimensional state space, realizing the quantitative representation of the state in all dimensions. This provides a complete input basis for subsequent adaptive alternating decision-making, ensuring the comprehensiveness of perception of complex dynamic environments and the consistency of state representation.

[0024] S202. Input the multidimensional state representation observed in the current time slot in the multidimensional state space into the policy network obtained in advance through reinforcement learning for decision reasoning, and generate a multi-domain alternation policy set for the next time slot with the goal of optimizing task performance.

[0025] In this embodiment, the policy network used is a deep reinforcement learning network, comprising a state embedding layer, multiple fully connected hidden layers, and a policy output layer. The state embedding layer performs high-dimensional feature mapping on the multi-dimensional state representation of the input; the multiple hidden layers perform nonlinear feature transformation and logical reasoning; and the policy output layer outputs the mode selection probability and continuous configuration parameters for each technical domain. During the training phase, different low-altitude operation scenarios are simulated in a simulation environment. The overall operating effect of the closed-loop system is quantitatively evaluated using the closed-loop performance index as the core reward function. Specifically, the reward value is positively correlated with perception resolution, communication reliability, and computational real-time performance, and negatively correlated with system power consumption and transmission latency. Through massive trial-and-error iterative optimization of network parameters, the policy network learns the optimal alternating scheduling strategy under different states.

[0026] During the inference phase, the multi-dimensional state representation of the current time slot is input into the trained policy network. The policy network performs forward computation based on the learned network parameters, adaptively generating and outputting the optimal multi-domain alternation policy set for the next time slot. After policy generation, a lightweight air interface control signaling interaction module quickly synchronizes configuration parameters between the UAV and the edge server. Specifically, the edge side encapsulates the policy set into short frame control signaling and sends it to the UAV through the downlink control channel. After receiving the signal, the UAV parses the parameters, configures its local hardware and software modules, and returns an acknowledgment frame to the edge side, ensuring time slot alignment and state consistency during mode switching at both ends.

[0027] For example, in a financial insurance claims investigation scenario, the policy network outputs a multi-domain alternation strategy adapted to the current working conditions based on the complex urban building environment, channel fluctuation status, edge computing load, and the urgency of the claims investigation task. This balances operational efficiency and resource consumption while ensuring claims investigation accuracy. In a medical drug transportation scenario, the policy network outputs a corresponding alternation strategy based on the current spatial environment obstructed by tall buildings, fast channel fading status, edge computing load, and drug transportation priority. This prioritizes ensuring the reliability of the control link and the speed of obstacle avoidance response during the transportation process.

[0028] This embodiment achieves end-to-end automatic generation of multi-domain policies through a pre-trained reinforcement learning policy network. It transforms the complex multi-domain collaborative judgment that originally relied on human experience into an automated end-to-end decision network. It replaces the traditional manual fixed configuration operation mode with overall task performance as the unified optimization goal, and realizes the collaborative adaptive scheduling of multi-domain resources, thereby improving the response speed and adaptability of UAV flight control to dynamic environments.

[0029] S203. Based on the multi-domain alternation strategy set, the perception operation mode, communication transmission mode and calculation offloading mode of the UAV are controlled in a coordinated and dynamic alternation, respectively, to perceive the environment and output the corresponding calculation results.

[0030] In this embodiment, based on the generated multi-domain alternation strategy set, the corresponding mode switching and parameter adjustment are dynamically arranged and executed in the perception domain, communication transmission domain and computing offloading domain. By dynamically adjusting the mode alternation ratio of each domain, the optimal running point can be explored along the better performance trade-off boundary in the four-dimensional performance space composed of perception accuracy, communication reliability, computing latency and resource consumption, and adapted to different resource constraints and task requirements.

[0031] This means that the three technical domains operate in different modes alternately and execute synchronously at the same time slot granularity. Within each time slot, the mode switching control of the three stages of sensing and acquisition, data transmission, and inference computation is completed sequentially. The duration of each stage is dynamically adapted according to the current mode, and the total duration does not exceed the time slot length. The mode of the sensing domain determines the accuracy and volume of the sensing data; the waveform mode of the communication domain must match the transmission requirements of the sensing data; and the offloading mode of the computing domain must match the communication bandwidth and computing load. The three domains achieve parameter coordination through joint decision-making via a policy network. For example, based on a multi-domain alternation policy set, the sensing domain can be switched to a high-precision bistatic mode, the communication domain can be switched to a high-throughput transmission mode, and the computing domain can be adjusted synchronously to adapt to a larger data volume. In this dynamically alternating operation mode, the entire process of environmental perception, data transmission, and inference computation is completed sequentially, thereby perceiving the environment and outputting the inference computation results corresponding to the job requirements.

[0032] For example, in a financial insurance claims investigation scenario, when the task transitions to open areas with standard communication coverage, the perception accuracy level, transmission anti-interference mode, and computation offloading method are adjusted collaboratively according to the generated multi-domain alternation strategy set. For instance, switching to monostatic perception and overall offloading mode saves energy and improves the investigation coverage efficiency per unit time. In a medical drug transport scenario, when the task requires high-precision obstacle avoidance in unknown urban areas, the alternation frequency of bistatic perception and segmentation computation is automatically increased according to the generated multi-domain alternation strategy set. This ensures high-resolution perception and low-latency inference, maintaining control link stability while ensuring obstacle perception response speed, thus ensuring the safety of drug transport.

[0033] This embodiment breaks through the bottleneck of the traditional single fixed operation mode by coordinating and dynamically alternating the control of the three core domains of perception, communication and computing, thereby achieving a dynamic balance of multi-dimensional performance.

[0034] S204. Based on the calculation results, control the UAV to perform corresponding actions in the next time slot, continue to acquire and process new multidimensional state representations as multidimensional state representations for the next time slot, and update the multidimensional state space.

[0035] In this embodiment, the calculation results obtained by perception computing include obstacle detection results, environmental semantic segmentation map, optimal flight path, equipment operating parameters and other information. Based on the calculation results, corresponding flight control commands are generated and sent to the UAV flight control system to drive the UAV to perform corresponding flight actions in the next time slot, thereby adjusting the flight attitude and flight path, and completing tasks such as obstacle avoidance, cruise, and fixed-point observation.

[0036] After the UAV completes its maneuver, its spatial position, motion state, and surrounding environment all change. Therefore, a new round of multi-source state data acquisition and standardization processing is initiated to generate a new multi-dimensional state representation for the next time slot and update the multi-dimensional state space, automatically entering the next round of closed-loop iteration. That is, using the multi-source state representation of the next time slot as input, the above-mentioned processes of state acquisition, strategy decision-making, alternating orchestration, and control issuance are executed cyclically to complete continuous adaptive closed-loop orchestration and flight control, forming a continuously iterative closed-loop operation mechanism.

[0037] For example, in a financial insurance claims investigation scenario, based on the calculation results of environmental perception and disaster loss identification, the drone is controlled to adjust its flight path to cover more areas to be investigated, while simultaneously collecting status data from the new location to update the state space, entering the next round of closed-loop control for investigation. In a medical drug transport scenario, based on the calculation results of obstacle perception and path planning, the drone is controlled to adjust its flight attitude to avoid building obstacles, while simultaneously updating the current airspace status data to enter the next round of closed-loop control, ensuring continuous obstacle avoidance throughout the transport process.

[0038] In the above embodiments, this invention discloses an adaptive flight control method based on dynamic alternating orchestration. It constructs a multi-dimensional state space by continuously collecting multi-source state data from the operational scenario and standardizing it into multi-dimensional state representations. The state representation of the current time slot is input into a pre-trained reinforcement learning policy network to generate a multi-domain alternating policy set for the next time slot with the goal of optimizing task performance. Based on the policy set, collaborative dynamic alternating control is performed on the perception operation mode, communication transmission mode, and computation offloading mode to complete environmental perception and inference calculations and output the calculation results. Based on the calculation results, the UAV is controlled to perform corresponding actions, synchronously updating the state space of the next time slot. This invention can be applied to business scenarios such as fintech and healthcare, realizing adaptive collaborative scheduling of multi-domain resources, effectively adapting to complex dynamic low-altitude environments, and improving the environmental adaptability and operational stability of UAV flight control.

[0039] In one embodiment, such as Figure 3As shown, step S201 includes: S301. Continuously acquire environmental status data, channel quality data, computing load data, and task level data of the drone in the operation scenario; S302. Standardize and encode the environmental state data, channel quality data, computing load data, and task level data to output a unified format multidimensional state representation and construct the multidimensional state space.

[0040] In this embodiment, the multi-source state data specifically includes four dimensions to comprehensively cover the key factors affecting the closed-loop operation effect. Among them, environmental state data is collected by UAV onboard sensors, reflecting information such as obstacle distribution, electromagnetic environment, and terrain in the operational airspace; channel quality data is obtained by real-time monitoring of the air-to-ground communication link, including link state indicators such as signal strength, signal-to-noise ratio, and Doppler shift estimation; computing power load data is reported in real-time by edge servers and cloud nodes, including computing power operation indicators such as processor utilization, memory usage, and task queuing latency; and task level data is issued by the task management system, representing the urgency and priority requirements of the current operation task.

[0041] After acquiring the above four types of data, a unified standardized encoding process is performed. For numerical data, max-min normalization or Z-score normalization can be used to eliminate dimensional differences. For categorical data, one-hot encoding or embedding encoding is used to convert it into vector form. Finally, all dimensions of data are concatenated into a fixed-length one-dimensional vector, i.e., a multidimensional state representation. Using this state representation as the basic unit, a continuous multidimensional state space is constructed by accumulating over time.

[0042] This embodiment acquires four core types of state data—environment, channel, computing power, and task—and performs standardized encoding processing to achieve a unified representation of heterogeneous data and construct a comprehensive multidimensional state space. This provides accurate and complete state input for subsequent reinforcement learning decisions, ensuring the accuracy and reliability of alternating strategy decisions.

[0043] In one embodiment, such as Figure 4 As shown, the multi-domain alternation strategy set includes a sensing mode selection instruction, an anti-Doppler waveform alternation instruction, and calculation configuration parameters. Step S203 includes: S401. According to the perception mode selection instruction, the system alternates between airborne monostatic mode and network-assisted bistatic mode in a time-division manner to acquire perception imaging data of corresponding accuracy. S402. According to the anti-Doppler waveform alternation instruction, dynamically switch the transmission waveform mode in the uplink sensing data transmission link, perform corresponding modulation processing on the sensing imaging data, and output the modulated uplink sensing data stream. S403. Dynamically adjust the running ratio of different computing modes according to the computing configuration parameters, perform inference calculation on the uplink sensing data stream, and output the corresponding calculation results.

[0044] In this embodiment, the multi-domain alternation strategy set specifically includes three types of executable instruction parameters, which correspond to the alternation control of the three technical domains of sensing, communication, and computing, respectively. The three are executed synchronously in the same time slot to ensure the coordination of mode switching and link matching. Among them, the sensing mode selection instruction is used to determine the sensing operation architecture of the next time slot, realizing a dynamic trade-off between sensing accuracy and airborne power consumption; the anti-Doppler waveform alternation instruction is used to determine the waveform format of uplink data transmission, realizing a balance between communication reliability and receiver computational complexity; and the computing configuration parameters are used to determine the offloading method of computing tasks and the network slicing depth, thereby dynamically balancing computing real-time performance and transmission bandwidth usage.

[0045] Specifically, the three types of instructions take effect synchronously within the same time slot. During execution, the instruction is first selected based on the perception mode, and the system alternates between airborne monostatic mode and network-assisted bistatic mode to complete environmental perception in the current time slot and generate perception imaging data of corresponding accuracy. Then, based on the anti-Doppler waveform alternation instruction, the transmission waveform mode is dynamically switched to modulate the perception imaging data, and the modulated uplink perception data stream is transmitted to the computing side through the air-to-ground link. Finally, based on the computing configuration parameters, the corresponding computing mode is selected to complete the inference calculation of the uplink perception data stream and output the final calculation result. The three stages are connected to form a coherent alternating execution link.

[0046] Furthermore, since the reinforcement learning policy network automatically completes the joint optimization of the alternating arrangement of the three domain modes during the decision-making phase, the mode selection of the three domains can be coordinated and matched. For example, when the task requires high-precision perception, the policy network will simultaneously select the bistatic perception mode, the high-reliability transmission mode, and the segmented computation mode, and the three will work together to ensure the acquisition, transmission, and processing of high-precision data. When the system is in a low-power cruise state, the policy network will simultaneously select the monostatic perception mode, the low-complexity transmission mode, and the local computation mode, and the three will work together to reduce the overall power consumption of the system.

[0047] This embodiment decomposes the multi-domain alternation strategy into three types of execution instructions, realizing time-sharing coordinated alternation control of the three domains of perception, communication, and computing. Through the dynamic combination of the three types of instructions, a continuously adjustable performance balance space is formed, and the adaptability to different working conditions is effectively improved.

[0048] In one embodiment, such as Figure 5 As shown, step S401 includes: S501. When it is determined that the current airborne monostatic mode is being executed, the communication waveform is multiplexed as a radar probe, and the UAV collects and performs preliminary imaging processing on the echo signal, and outputs local perception imaging data. S502. When it is determined that the current execution is in network-assisted bistatic mode, the UAV is controlled as a mobile transmitter to output detection signals and communication coverage signals. The ground base station captures the echo signals and performs backward projection imaging processing to output network perception imaging data.

[0049] In this embodiment, the airborne monostatic mode refers to the integrated sensing and communication mode in which both the transmitting and receiving ends of radar detection are deployed on the UAV, and it has autonomous sensing capabilities; the network-assisted bistatic mode refers to the integrated sensing and communication mode in which the transmitting end is deployed on the UAV and the receiving end and imaging processing end are deployed on the ground base station, which can achieve high-precision imaging.

[0050] During continuous flight time slots, the UAV alternates between airborne monostatic mode and network-assisted bistatic mode based on perception mode selection commands. In airborne monostatic mode, the UAV simultaneously performs signal transmission and echo reception, multiplexing the communication signal waveform into a radar detection signal for transmission, while simultaneously capturing echo signals reflected from the environment via the airborne receiving antenna. The airborne processing unit performs echo preprocessing and preliminary imaging algorithms to generate local perception imaging data. This mode does not rely on the computing resources of ground base stations and can still autonomously complete environmental perception even when the link is interrupted, ensuring basic obstacle avoidance requirements. However, due to limitations in airborne computing power, the imaging resolution of the local perception imaging data is relatively low, and power consumption is relatively high.

[0051] In the network-assisted bistatic mode, the transceiver is deployed separately. The UAV acts only as a mobile transmitter, continuously emitting integrated detection and communication signals. The echo signals are captured with high gain by the base station antenna deployed on the ground, and then the computing unit on the base station side executes a high-precision back projection imaging algorithm. Since the back projection algorithm can directly perform point-by-point coherent accumulation in the imaging space, it is not limited by the bistatic geometry, resulting in higher imaging accuracy and smaller geometric distortion, thus outputting higher-resolution network-sensing imaging data. This mode offloads the heavy imaging computation to the network side, but it has a large computational load and high power consumption.

[0052] By adjusting the proportion of the two modes within a continuous time slot, a continuous trade-off between sensing accuracy and system power consumption is achieved. When high-precision environmental modeling is required, the proportion of the bistatic mode is increased; when power is tight or the link is unstable, the proportion of the monostatic mode is increased, thereby achieving optimal sensing performance under the constraints of airborne size, weight, and power consumption.

[0053] This embodiment achieves a time-sharing alternation of monostatic and bistatic sensing modes, ensuring sensing autonomy in extreme scenarios while maintaining high-precision sensing capabilities when the link is stable. By dynamically alternating the ratio, a balance is achieved between sensing performance and airborne resource consumption, effectively alleviating the contradiction between UAV hardware constraints and the need for high-precision sensing.

[0054] In one embodiment, such as Figure 6 As shown, step S402 includes: S601. Real-time detection of Doppler frequency shift estimate. When the Doppler frequency shift estimate is greater than a preset threshold, switch to a radio frequency division multiplexing transmission mode in the uplink sensing data transmission link, modulate the sensing imaging data and output the uplink sensing data stream. S602. When the Doppler frequency shift estimate is less than or equal to a preset threshold, the uplink sensing data transmission link is switched to an orthogonal frequency division multiplexing transmission mode that mimics radio frequency domain pilots, and the sensing imaging data is modulated and then the uplink sensing data stream is output.

[0055] In this embodiment, in complex low-altitude scenarios, the drastic changes in the UAV's flight speed and heading angle cause frequent switching between slow and fast fading in the channel. Therefore, an anti-Doppler waveform alternation command is used to dynamically alternate the two uplink transmission waveform modes under different Doppler effects, thus balancing anti-Doppler performance with receiver computational complexity. Specifically, the degree of time-varying drasticness of the current channel can be reflected by the Doppler frequency shift estimate, which is determined by the degree of frequency shift of the received signal caused by the UAV's high-speed movement. A larger value indicates more drastic time-varying channel drasticness.

[0056] When the Doppler shift estimate exceeds a preset threshold, it indicates that the channel is in a high-speed, fast-fading state. The specific preset threshold can be pre-calibrated according to the system's bit error rate requirements. When the Doppler shift exceeds the threshold, the inter-carrier interference of traditional orthogonal frequency division multiplexing will increase significantly, and the bit error rate will exceed the acceptable range. Therefore, at this time, the transmission mode is switched to affine radio frequency division multiplexing. Based on the affine Fourier transform, a multi-carrier signal is constructed, and its full diversity gain is used to combat the fading of the dual-selection channel, ensuring the transmission reliability in high-speed mobile scenarios.

[0057] When the Doppler frequency shift estimate is below or equal to a preset threshold, it indicates that the channel is relatively stable. At this point, the transmission mode switches to an orthogonal frequency division multiplexing (OFDM) transmission mode assisted by pilots in the anamorphic radio frequency (RF) domain. Data transmission in this mode follows the time-frequency domain architecture of OFDM, only mapping the pilot signals to the anamorphic radio frequency domain via affine Fourier transform. That is, high-precision channel estimation is first performed in the anamorphic radio frequency domain using pilots, and then the channel state information is converted back to the time-frequency domain for conventional frequency domain equalization of the data symbols. This communication mode can obtain high-precision channel state information by leveraging the sparsity of the anamorphic radio frequency domain, while retaining the mature, low-complexity equalization architecture of OFDM, thus significantly reducing the computational overhead at the receiver while maintaining flexible time-frequency resource scheduling capabilities.

[0058] In addition, for downlink control signaling transmission, a symbol-level precoding mechanism with pre-equalization at the transmitter can be used to transfer channel estimation and equalization operations from the computing-limited UAV receiver to the edge transmitter by utilizing channel reciprocity. The edge server completes the precoding process based on the channel estimation results of the simulated radio frequency domain pilot. The UAV only needs low-complexity direct symbol detection to receive control commands, further ensuring the reliability of control command transmission in high-mobility scenarios.

[0059] This embodiment achieves a dynamic balance between communication performance and computational overhead by dynamically alternating between two uplink transmission modes based on the Doppler state. This ensures transmission reliability in high Doppler scenarios, reduces reception complexity and improves scheduling flexibility in stable channels, and thus adapts to the rapid channel changes caused by the high-speed maneuvering of low-altitude UAVs.

[0060] In one embodiment, such as Figure 7 As shown, step S403 includes: S701. Obtain the calculation unloading ratio and the segmentation point from the calculation configuration parameters; S702. Dynamically adjust the ratio of the number of runs between the overall unloading calculation mode and the cross-layer segmentation calculation mode according to the calculated unloading ratio, and perform network slicing in the cross-layer segmentation calculation mode according to the segmentation point. S703. Perform inference calculations on the uplink sensing data stream under the current dynamically determined calculation mode, and output the corresponding calculation results.

[0061] In this embodiment, the computing side is configured with two modes: overall offloading computing and cross-layer segmentation computing. Overall offloading computing mode treats the complete perceptual reasoning task as an indivisible whole and distributes it to a single computing node, such as local, edge, or cloud, for execution. Cross-layer segmentation computing mode divides the deep neural network model into layers and distributes them across multiple computing nodes for collaborative execution, significantly reducing the amount of data transmitted. The proportion of operation for the two modes is dynamically adjusted by obtaining the offloading ratio parameter in the computing configuration parameters, and the network layer for cross-layer segmentation is controlled by obtaining the segmentation point parameter in the computing configuration parameters, achieving a dynamic trade-off between computing latency and transmission bandwidth.

[0062] The computational configuration parameters include two core parameters: the computational offloading ratio and the segmentation point. The computational offloading ratio is a continuous value between 0 and 1, representing the proportion of time slots occupied by the cross-layer segmentation computation mode within multiple consecutive time slots. A higher ratio indicates a greater tendency to use the low-bandwidth transmission segmentation computation mode. The segmentation point parameter specifies the specific segmentation layer number of the neural network model. For example, in a feature extraction network containing 8 convolutional blocks, a segmentation point value of 2 indicates that slicing occurs after the second convolutional block.

[0063] In the overall offload computing mode, based on the granularity of the perception task, latency requirements, and the load status of each computing node, a scheduling algorithm distributes the complete inference task to a single node in the drone's local machine, edge server, or cloud for execution, outputting complete inference results. For example, small-granularity, low-latency tasks are assigned to local or edge execution, while large-granularity, non-real-time tasks are assigned to the cloud. This computing mode has simple task scheduling logic and guaranteed inference accuracy, making it suitable for scenarios with ample bandwidth and low computing load.

[0064] In the cross-layer segmentation computation mode, based on the hierarchical structure of deep neural networks, the inference task is divided into multiple parts by slicing at specified network layers according to the segmentation point parameters. These parts are then executed collaboratively on the drone, edge, and cloud. The drone only executes the shallow network before the segmentation point to complete the front-end feature extraction of the raw perception data and outputs a highly compressed intermediate feature map to the edge server. The edge server then executes the mid-layer network after the segmentation point to complete semantic inference such as object detection and semantic segmentation, outputting semantic-level intermediate results. These semantic-level intermediate results are further uploaded to the cloud, where deep tasks such as global optimization, path planning, and long-term decision-making are performed to output the final refined decision computation results.

[0065] Preferably, to avoid semantic fragmentation and accuracy loss caused by network slicing, a multimodal feature alignment network based on attention mechanism is further integrated. This network is deployed at the inference entry point at the edge and cloud. Through channel attention and spatial attention mechanisms, the intermediate features transmitted from the front end are calibrated and aligned to ensure that they are consistent with the input distribution of the back end network, thereby compensating for the feature distribution offset caused by slicing and ensuring that the inference accuracy is not affected by segmentation calculation.

[0066] This embodiment achieves hierarchical and flexible orchestration of computing tasks by dynamically adjusting the operating ratio of the two computing modes and the network segmentation point. This allows the task to automatically alternate between segmented computing when bandwidth is sharply reduced, transmitting only semantic features at extremely low bit rates. When bandwidth is restored, the task quickly alternates back to overall offloading to obtain more comprehensive environmental information. This effectively alleviates the contradiction between air-to-ground link bandwidth constraints and high-precision inference requirements, ensuring that end-to-end inference latency remains within the allowable closed-loop response range even in extremely resource-constrained environments, thus guaranteeing the real-time performance of end-to-end inference.

[0067] In one embodiment, such as Figure 8 As shown, step S204 includes: S801. Generate corresponding flight control commands based on the calculation results and send them to the UAV; S802. Control the UAV to adjust its flight attitude and flight path according to the flight control command at the next moment in order to complete the corresponding action.

[0068] In this embodiment, the calculation results include information such as environmental perception and recognition results, obstacle distribution, and path planning scheme. Based on this information, corresponding flight control commands are generated, covering control quantities such as attitude adjustment, speed change, and path correction. These commands are sent to the UAV flight control system through the downlink communication link. Specifically, a short frame retransmission mechanism can be used to ensure transmission reliability and avoid flight loss due to command loss.

[0069] After receiving control commands, the UAV flight control system drives the UAV's power system and actuators to adjust its flight attitude, speed, and path, achieving precise adjustments to the flight attitude and path to reliably complete corresponding operational actions such as obstacle avoidance, cruising, and fixed-point observation. This embodiment generates precise control commands based on calculation results and drives the UAV to execute actions, ensuring the continuity of UAV operations and control accuracy in dynamic environments.

[0070] It should be noted that there is no necessary order between the above steps. Those skilled in the art will understand from the description of the embodiments of the present invention that the above steps may have different execution orders in different embodiments, that is, they may be executed in parallel or in turn, etc.

[0071] Further reference Figure 9 As a response to the above Figure 2 The present invention provides an embodiment of an adaptive flight control device based on dynamic alternating arrangement, which is similar to the method shown. Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0072] like Figure 9 As shown, the adaptive flight control device 90 based on dynamic alternating arrangement described in this embodiment includes: The state space construction module 901 is used to continuously acquire multi-source state data of the UAV in the operation scenario and perform standardized processing, and construct the corresponding multi-dimensional state space based on the processed multi-dimensional state representation. The strategy decision module 902 is used to input the multidimensional state representation observed in the current time slot in the multidimensional state space into the policy network obtained in advance through reinforcement learning for decision reasoning, and generate a multi-domain alternating policy set for the next time slot with the goal of optimizing task performance. The dynamic alternation orchestration module 903 is used to perform coordinated dynamic alternation control of the UAV's perception operation mode, communication transmission mode and computing offload mode according to the multi-domain alternation strategy set, perceive the environment and output the corresponding calculation results. The control execution module 904 is used to control the UAV to perform corresponding actions in the next time slot according to the calculation results, continue to acquire and process new multidimensional state representations as multidimensional state representations for the next time slot, and update the multidimensional state space.

[0073] The module referred to in this invention is a series of computer program instruction segments that can perform specific functions. It is more suitable than a program for describing the execution process of adaptive flight control based on dynamic alternation. For specific implementation methods of each module, please refer to the corresponding method embodiments above, which will not be repeated here.

[0074] In one embodiment, the state space construction module 901 includes: The data acquisition unit is used to continuously acquire environmental status data, channel quality data, computational load data, and task level data of the UAV in the operational scenario. The standardization processing unit is used to perform standardized encoding processing on the environmental state data, channel quality data, computational load data, and task level data, output a unified format multidimensional state representation, and construct the multidimensional state space.

[0075] In one embodiment, the multi-domain alternation strategy set includes a sensing mode selection instruction, an anti-Doppler waveform alternation instruction, and calculation configuration parameters; the dynamic alternation orchestration module 903 includes: The perception mode switching unit is used to perform time-division alternation between airborne monostatic mode and network-assisted bistatic mode according to the perception mode selection instruction, so as to acquire perception imaging data of corresponding accuracy. The communication mode switching unit is used to dynamically switch the transmission waveform mode in the uplink sensing data transmission link according to the anti-Doppler waveform alternation instruction, perform corresponding modulation processing on the sensing imaging data, and output the modulated uplink sensing data stream. The computing mode switching unit is used to dynamically adjust the running ratio of different computing modes according to the computing configuration parameters, perform inference calculations on the uplink sensing data stream, and output the corresponding calculation results.

[0076] In one embodiment, the sensing mode switching unit is specifically used for: When it is determined that the current airborne monostatic mode is being executed, the communication waveform is multiplexed as a radar probe, and the UAV collects and performs preliminary imaging processing on the echo signal to output local perception imaging data. When it is determined that the current execution mode is network-assisted bistatic mode, the UAV is controlled as a mobile transmitter to output detection signals and communication coverage signals. The ground base station captures the echo signals and performs back projection imaging processing to output network perception imaging data.

[0077] In one embodiment, the communication mode switching unit is specifically used for: The Doppler frequency shift estimate is detected in real time. When the Doppler frequency shift estimate is greater than a preset threshold, the transmission mode is switched to simulated radio frequency multiplexing in the uplink sensing data transmission link. The sensing imaging data is modulated and then the uplink sensing data stream is output. When the Doppler frequency shift estimate is less than or equal to a preset threshold, the uplink sensing data transmission link is switched to the orthogonal frequency division multiplexing transmission mode of the radio frequency domain pilot, and the sensing imaging data is modulated and then the uplink sensing data stream is output.

[0078] In one embodiment, the computing mode switching unit is specifically used for: Obtain the calculation unloading ratio and the segmentation point from the calculation configuration parameters; The ratio of the number of runs of the overall unloading calculation mode and the cross-layer segmentation calculation mode is dynamically adjusted according to the calculated unloading ratio, and network slicing is performed in the cross-layer segmentation calculation mode according to the segmentation point. In the current dynamically determined computing mode, inference calculations are performed on the uplink sensing data stream, and the corresponding calculation results are output.

[0079] In one embodiment, the control execution module 904 includes: The instruction generation and distribution unit is used to generate corresponding flight control instructions based on the calculation results and distribute them to the UAV. The execution unit is used to control the UAV to adjust its flight attitude and flight path in the next time slot according to the flight control command, so as to complete the corresponding action.

[0080] In the above embodiments, this invention discloses an adaptive flight control device based on dynamic alternating orchestration. It continuously collects multi-source state data from various operational scenarios and standardizes it into multi-dimensional state representations to construct a multi-dimensional state space. The state representation of the current time slot is input into a pre-trained reinforcement learning policy network to generate a multi-domain alternating policy set for the next time slot, aiming for optimal task performance. Based on the policy set, it performs coordinated dynamic alternating control on the perception operation mode, communication transmission mode, and computation offloading mode, completing environmental perception and inference calculations and outputting the calculation results. Based on the calculation results, it controls the UAV to perform corresponding actions, synchronously updating the state space for the next time slot. This invention can be applied to business scenarios such as fintech and healthcare, achieving adaptive coordinated scheduling of multi-domain resources, effectively adapting to complex dynamic low-altitude environments, and improving the environmental adaptability and operational stability of UAV flight control.

[0081] Specific limitations regarding the adaptive flight control device based on dynamic alternating staging can be found in the limitations of the adaptive flight control method based on dynamic alternating staging described above, and will not be repeated here. Each module in the aforementioned adaptive flight control device based on dynamic alternating staging can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0082] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0083] Another embodiment of the present invention provides a computer device, such as... Figure 10 As shown, the computer device 100 includes: One or more processors 1001 and memory 1002, Figure 10 The following description uses a processor 1001 as an example. The processor 1001 and the memory 1002 can be connected via a bus or other means. Figure 10 Taking the example of a connection between China and Israel via a bus.

[0084] The processor 1001 is used to perform various control logics of the computer device 100. It can be any conventional processor, microprocessor, state machine, general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), microcontroller, ARM (Acorn RISC Machine) or other programmable logic device, discrete gate or transistor logic, discrete hardware components or any combination of these components.

[0085] The memory 1002, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions corresponding to the adaptive flight control method based on dynamic alternating arrangement in the embodiments of the present invention. The processor 1001 executes various functional applications and data processing of the computer device 100 by running the non-volatile software programs, instructions, and units stored in the memory 1002, thereby implementing the adaptive flight control method based on dynamic alternating arrangement in the above method embodiments.

[0086] Another embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by one or more processors, perform the steps of the adaptive flight control method based on dynamic alternation arrangement in any of the above method embodiments.

[0087] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0088] Based on the above description of the embodiments, those skilled in the art will understand that the methods described in the embodiments can be implemented using software plus necessary general-purpose hardware platforms. Of course, they can also be implemented using hardware, but in many cases, the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0089] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The computer program can be stored in a non-volatile, computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The storage medium can be a memory, magnetic disk, floppy disk, flash memory, optical storage, etc.

[0090] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0091] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An adaptive flight control method based on dynamic alternating arrangement, characterized in that, include: Continuously acquire multi-source state data of drones in operational scenarios and perform standardized processing; construct a corresponding multi-dimensional state space based on the processed multi-dimensional state representation. The multidimensional state representation observed in the current time slot in the multidimensional state space is input into the policy network obtained in advance through reinforcement learning for decision reasoning, and a multi-domain alternating policy set for the next time slot is generated with the goal of optimizing task performance. Based on the multi-domain alternation strategy set, the perception operation mode, communication transmission mode and computation offloading mode of the UAV are controlled in a coordinated and dynamic alternation, respectively, to perceive the environment and output the corresponding calculation results. Based on the calculation results, the UAV is controlled to perform corresponding actions in the next time slot, and new multidimensional state representations are obtained and processed as multidimensional state representations for the next time slot, and the multidimensional state space is updated.

2. The adaptive flight control method based on dynamic alternating arrangement according to claim 1, characterized in that, The process of continuously acquiring multi-source state data of the UAV in the operational scenario and performing standardized processing, and constructing a corresponding multi-dimensional state space based on the processed multi-dimensional state representation, includes: Continuously acquire environmental status data, channel quality data, computational load data, and task level data of the drone in the operational scenario; The environmental state data, channel quality data, computational load data, and task level data are standardized and encoded to output a unified format of multidimensional state representation and construct the multidimensional state space.

3. The adaptive flight control method based on dynamic alternating arrangement according to claim 1, characterized in that, The multi-domain alternation strategy set includes perception mode selection instructions, anti-Doppler waveform alternation instructions, and calculation configuration parameters. Based on the multi-domain alternation strategy set, the UAV's perception operation mode, communication transmission mode, and calculation offloading mode are controlled in a coordinated and dynamic alternation manner to perceive current environmental information and output corresponding calculation results, including: According to the perception mode selection instruction, the system alternates between airborne monostatic mode and network-assisted bistatic mode in a time-division manner to acquire perception imaging data of corresponding accuracy. According to the anti-Doppler waveform alternation command, the transmission waveform mode is dynamically switched in the uplink sensing data transmission link to perform corresponding modulation processing on the sensing imaging data and output the modulated uplink sensing data stream. The proportion of operation of different computing modes is dynamically adjusted according to the computing configuration parameters, and inference calculations are performed on the uplink sensing data stream to output the corresponding calculation results.

4. The adaptive flight control method based on dynamic alternating arrangement according to claim 3, characterized in that, The step of alternating between airborne monostatic mode and network-assisted bistatic mode according to the sensing mode selection instruction to acquire sensing imaging data of corresponding accuracy includes: When it is determined that the current airborne monostatic mode is being executed, the communication waveform is multiplexed as a radar probe, and the UAV collects and performs preliminary imaging processing on the echo signal to output local perception imaging data. When it is determined that the current execution mode is network-assisted bistatic mode, the UAV is controlled as a mobile transmitter to output detection signals and communication coverage signals. The ground base station captures the echo signals and performs back projection imaging processing to output network perception imaging data.

5. The adaptive flight control method based on dynamic alternating arrangement according to claim 3, characterized in that, The step of dynamically switching the transmission waveform mode in the uplink sensing data transmission link according to the anti-Doppler waveform alternation command, performing corresponding modulation processing on the sensing imaging data, and outputting the modulated uplink sensing data stream includes: The Doppler frequency shift estimate is detected in real time. When the Doppler frequency shift estimate is greater than a preset threshold, the transmission mode is switched to simulated radio frequency multiplexing in the uplink sensing data transmission link. The sensing imaging data is modulated and then the uplink sensing data stream is output. When the Doppler frequency shift estimate is less than or equal to a preset threshold, the uplink sensing data transmission link is switched to the orthogonal frequency division multiplexing transmission mode of the radio frequency domain pilot, and the sensing imaging data is modulated and then the uplink sensing data stream is output.

6. The adaptive flight control method based on dynamic alternating arrangement according to claim 3, characterized in that, The process of dynamically adjusting the running ratio of different computing modes according to the computing configuration parameters, performing inference calculations on the uplink sensing data stream, and outputting corresponding calculation results includes: Obtain the calculation unloading ratio and the segmentation point from the calculation configuration parameters; The ratio of the number of runs of the overall unloading calculation mode and the cross-layer segmentation calculation mode is dynamically adjusted according to the calculated unloading ratio, and network slicing is performed in the cross-layer segmentation calculation mode according to the segmentation point. In the current dynamically determined computing mode, inference calculations are performed on the uplink sensing data stream, and the corresponding calculation results are output.

7. The adaptive flight control method based on dynamic alternating arrangement according to claim 1, characterized in that, The step of controlling the UAV to perform corresponding actions in the next time slot based on the calculation results includes: Based on the calculation results, corresponding flight control commands are generated and sent to the UAV; The drone is controlled to adjust its flight attitude and flight path in the next time slot according to the flight control command in order to complete the corresponding actions.

8. An adaptive flight control device based on dynamic alternating arrangement, characterized in that, include: The state space construction module is used to continuously acquire multi-source state data of UAVs in the operation scenario and perform standardized processing, and construct the corresponding multi-dimensional state space based on the processed multi-dimensional state representation. The strategy decision module is used to input the multidimensional state representation observed in the current time slot in the multidimensional state space into the policy network obtained in advance through reinforcement learning for decision reasoning, and generate a multi-domain alternating policy set for the next time slot with the goal of optimizing task performance. The dynamic alternation orchestration module is used to perform coordinated dynamic alternation control of the UAV's perception operation mode, communication transmission mode and computing offload mode according to the multi-domain alternation strategy set, perceive the environment and output the corresponding calculation results. The control execution module is used to control the UAV to perform corresponding actions in the next time slot according to the calculation results, continue to acquire and process new multidimensional state representations as multidimensional state representations for the next time slot, and update the multidimensional state space.

9. A computer device, characterized in that, Includes at least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the adaptive flight control method based on dynamic alternation as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when executed by one or more processors, cause the one or more processors to perform the adaptive flight control method based on dynamic alternation as described in any one of claims 1-7.