Low-altitude aircraft communication-restricted environment autonomous endurance and intelligent return system and method
By controlling the synchronous determination of the cycle, smooth attitude transition, and generalized airflow observation, combined with multi-source perception fusion positioning and lightweight federated learning, the problem of safe return of low-altitude aircraft in communication-limited environments has been solved, achieving efficient and safe autonomous endurance and intelligent return.
Patent Information
- Application Number
- CN202610786341.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-02
- Publication Date
- 2026-08-25
AI Technical Summary
Existing low-altitude aircraft face safety issues in complex scenarios such as communication restrictions and positioning failures, including disconnection between communication loss determination and flight control cycle, sudden attitude changes, and hard switching. Furthermore, existing technologies have a narrow scope of protection, making mass production and rights protection difficult.
It adopts an autonomous endurance and intelligent return system with control cycle synchronous determination, smooth attitude transition, generalized airflow observation, and lightweight self-evolution. Through multi-source perception fusion positioning, dynamic index lookup table, and lightweight anti-poisoning federated learning, it achieves bumpless switching and safe return.
It achieves high-frequency real-time control, wide adaptability, and safe return to base, adapts to both high- and low-cost hardware, has a creative protection scope, reduces the difficulty of rights protection, and ensures flight safety.
Smart Images
Figure CN122632899A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of autonomous control, endurance optimization, safe navigation and intelligent emergency technology for low-altitude aircraft. Specifically, it relates to an autonomous endurance and intelligent return system and method for low-altitude aircraft in scenarios with limited communication and positioning failure, which is based on control cycle synchronization determination, smooth attitude transition, generalized airflow observation and lightweight self-evolution. It is applicable to all types of low-altitude flight equipment such as logistics drones, manned eVTOL, inspection aircraft, and emergency rescue equipment. Background Technology
[0002] Low-altitude aircraft are highly susceptible to communication limitations, positioning failures, and sudden battery drains when operating in complex environments such as urban buildings obstructing traffic, mountain tunnels, areas with electromagnetic interference, and remote airspace. Existing technologies generally suffer from the following insurmountable technical defects:
[0003] The judgment of communication loss is completely out of sync with the flight control cycle. It only uses a fixed timeout judgment and is not synchronized with the high-frequency control cycle. It is easy to accidentally trigger the autonomous system in the control loop oscillation, which can lead to sudden changes in the aircraft's attitude, loss of control, or even crash.
[0004] The specific algorithms for airflow inversion and energy consumption optimization, such as Kalman filtering, have a narrow scope of protection, and competitors can easily circumvent patent protection by replacing the algorithms.
[0005] The hardware of the perception and positioning solution is too tightly bound, relying on a single radar or visual sensor, lacking a general upper-level concept, making it impossible to adapt and implement in low-cost models, and making it difficult to protect patent rights.
[0006] The federal learning security mechanism is not sufficiently transparent, and its data is rigid and lacks sufficient support in terms of lightweight implementation, making it easy to raise questions about its "infeasibility" or "lack of technical basis."
[0007] There is no smooth transition mechanism when switching between cloud control and local autonomy. The power output is discontinuous at the moment of switching, resulting in extremely poor flight safety.
[0008] Patent drafting is too concrete, with parameters, algorithms, and hardware all fixed and limited, making it easy to circumvent with equivalent substitutions and failing to form an effective patent barrier.
[0009] In summary, existing technologies cannot meet the requirements of low-altitude safe flight, which require high-frequency real-time control, wide adaptability, difficulty in avoidance, mass production, and strong protection. There is an urgent need for a complete closed-loop patented technology solution that is highly innovative and has a broad protection scope. Summary of the Invention
[0010] Purpose of the invention
[0011] This invention addresses the shortcomings of existing technologies by providing an autonomous endurance and intelligent return-to-home system and method for low-altitude aircraft in communication-constrained environments, achieving the following core objectives:
[0012] It solves the problems of misjudgment and attitude change caused by the asynchrony between communication judgment and control cycle, and achieves smooth and autonomous switching without disturbance, forming an innovation that is fundamentally different from existing technologies;
[0013] By using functional overarching limitations to replace specific algorithms, fixed parameters, and specific hardware, the scope of patent protection is expanded, and equivalent substitution is prevented from circumventing the limitations.
[0014] A universal multi-source perception definition, compatible with high-end radar and low-cost sensors, enables lightweight deployment across all models;
[0015] The federal learning program is lightweight, de-numerical, and feasible, meeting the requirements for full disclosure in patent examination.
[0016] Construct a dual patent protection system of "system + method" to cover all scenarios of hardware, software, algorithms and integration, and reduce the difficulty of rights protection;
[0017] This can be achieved in both high-end and low-end models, without relying on expensive hardware, and meets the compatibility requirements for mass production and patent examination.
[0018] Technical solution
[0019] An autonomous endurance and intelligent return system for low-altitude aircraft in communication-constrained environments includes a flight control main control unit (1), an endurance monitoring unit (2), a multi-source positioning unit (3), an environmental perception unit (4), a distributed communication unit (5), an execution unit (6), and a non-volatile storage unit (7); each unit is electrically connected to the flight control main control unit (1).
[0020] The multi-source positioning unit (3) includes an inertial navigation module (31), a visual SLAM module (32), an active spatial perception matching module (33), and a geomagnetic sensor; the environmental perception unit (4) includes an airflow inversion module (41) and a downward perception module (42); the non-volatile storage unit (7) carries a model self-evolution module (71).
[0021] 1. Communication limitation determination and smooth handover
[0022] Using the flight control cycle as the minimum time reference, uplink communication commands are continuously monitored at a periodic level. If no valid uplink commands are received for several consecutive control cycles, communication is deemed restricted, and the system automatically switches to local autonomous mode. The switching process executes an attitude smooth transition algorithm: at the moment of switching, the current angular velocity and attitude angle are locked as initial boundary conditions. Using a time constant τ or a preset gradual curve, the control weights are linearly or nonlinearly transferred from uplink commands to local autonomous commands within N control cycles, ensuring that the first derivative of the power output is continuous, without abrupt changes, oscillations, or instability, fundamentally different from existing timeout hard-cut return-to-home schemes.
[0023] 2. Generalized airflow state observer
[0024] A state observer is constructed based on the deviation between the aircraft dynamics model and sensor data to invert the forward airflow vector. The observer includes, but is not limited to, linear observers, Kalman-type observers, nonlinear observers, and neural network observers. Low-end computing power configurations employ a dynamic index lookup method, dynamically indexing a pre-stored drag coefficient matrix based on real-time acquired IMU vibration spectrum characteristics and air pressure change rate, adapting to drag characteristics under different flight attitudes, rather than using static table lookup.
[0025] 3. Universal multi-source fusion positioning
[0026] Based on inertial navigation, it integrates visual SLAM, active spatial perception, and geomagnetic heading to achieve joint positioning; active spatial perception includes integrated data from millimeter-wave radar, lidar, ultrasonic radar, and ISAC communication perception; when any perception unit fails, it automatically degrades to the remaining available perception combination to maintain continuous and uninterrupted navigation.
[0027] 4. Lightweight Federated Learning for Preventing Poisoning
[0028] On the local end, lightweight incremental training is achieved using knowledge distillation, model pruning, and model quantization. After communication is restored, only the model gradients are uploaded, not the original data. Before cloud aggregation, anomaly detection is performed based on the directional consistency or distribution statistical characteristics of the gradient vectors. Malicious gradients are removed, and then federated aggregation is completed to generate a global model and distribute it synchronously. This enables edge-cloud collaborative computing while taking into account onboard computing power limitations, data compliance, and model security.
[0029] 5. Dynamic return to base and tiered emergency response
[0030] The return path does not depend on the original flight route. It dynamically plans the path with the best energy consumption, fewest obstacles, and highest safety. It is divided into four levels according to the remaining power: normal, low power, ultra-low power, and emergency power. Low power triggers the return, ultra-low power initiates a safe emergency landing at the nearest safe location, and emergency power executes a protective touchdown.
[0031] 6. Distributed communication relay
[0032] It periodically scans ground-distributed nodes, automatically re-establishes communication links when the signal meets preset conditions, uploads flight status and receives control commands, thus meeting low-altitude monitoring requirements.
[0033] Creative distinction from existing technologies
[0034] Synchronous control cycle determination + continuous smooth switching of first derivative: Existing technology is fixed-duration timeout hard cutting, while this invention is high-frequency periodic soft cutting, with no attitude oscillation, which is an unexpected technical effect;
[0035] Generalized state observer: It only provides functional limitations, is not bound to a specific algorithm, and has a very wide protection range;
[0036] Dynamic index lookup method: It combines real-time operating conditions for adaptive adjustment, which is different from the known static lookup method and has innovation.
[0037] Gradient direction consistency + distribution statistics detection: rigorous theory, superior expression, no ambiguity, and full disclosure;
[0038] Dual protection of system and methodology: covering all scenarios of hardware, software, algorithms, and integration, making rights protection simple and efficient.
[0039] Beneficial effects
[0040] 1. Outstanding innovation: The communication cycle synchronization determination, smooth attitude transition, and continuous first derivative cannot be refuted by existing technologies;
[0041] 2. Broad protection scope: It removes parameterization, specific algorithms, and specific hardware, making it impossible to bypass with equivalent replacements;
[0042] 3. Lightweight and deployable: It can be implemented with both high and low computing power and high and low cost hardware, and is suitable for mass production needs;
[0043] 4. Full transparency: All functions provide theoretical logic and implementation paths, with no points for review or questioning;
[0044] 5. Simple rights protection: The methods and rights cover the entire process, and software, algorithms, and integration are all within the scope of protection;
[0045] 6. High safety: Seamless handover, failure detection and degradation, and safe emergency landing completely solve the problems of lost aircraft, crashes, and secondary damage. Attached Figure Description
[0046] Figure 1 is a schematic diagram of the overall hardware architecture of the system of the present invention;
[0047] Figure 2 is a schematic diagram of the autonomous endurance and intelligent return process of the present invention;
[0048] Figure 3 is a schematic diagram of the multi-source localization, emergency response and model self-evolution logic of the present invention.
[0049] The module consists of: 1-Flight Control Main Unit; 2-Endurance Monitoring Unit; 3-Multi-Source Positioning Unit; 31-Inertial Navigation Module; 32-Visual SLAM Module; 33-Active Spatial Perception Matching Module; 4-Environmental Perception Unit; 41-Airflow Inversion Module; 42-Look-Down Perception Module; 5-Distributed Communication Unit; 6-Execution Unit; 7-Non-Volatile Storage Unit; 71-Model Self-Evolution Module. Detailed Implementation
[0050] Implementation Method 1: Smooth Handover Due to Communication Constraints
[0051] The flight control cycle is the fundamental cycle for flight control attitude calculation and power output. The system continuously monitors the validity of uplink commands according to the control cycle. If no valid uplink commands are received for several consecutive control cycles, communication is deemed restricted and an autonomous handover is triggered. Attitude smooth transition algorithm: At the moment of handover, the current angular velocity and attitude angle are locked as initial boundary conditions. With time constant τ or a gradual curve as constraints, the control weights are smoothly transferred from uplink commands to local autonomous commands within N control cycles, ensuring the continuity of the first derivative of power output and that the attitude is free from sudden changes, oscillations, and instability.
[0052] Implementation Method 2: Generalized Airflow Observer and Dynamic Index Lookup Table
[0053] Based on the aircraft dynamics model, the deviation between the sensor's measured values and theoretical estimates is used as input to invert the forward airflow vector through a state observer. High-end configurations employ filtered or nonlinear observers, while low-end computing power configurations use a dynamic index lookup method. This dynamic index lookup method is not a static search; instead, it dynamically indexes a pre-stored drag coefficient matrix based on real-time IMU vibration spectrum characteristics and air pressure change rate, adaptively matching the drag characteristics corresponding to the current flight attitude, altitude, and speed, thus possessing adaptive operating condition capabilities.
[0054] Implementation Method 3: Generalized Multi-Source Fusion Positioning
[0055] Based on the inertial navigation module (31), the data of the visual SLAM module (32) and the active spatial perception matching module (33) are fused; the active spatial perception can use millimeter-wave radar, ultrasonic radar or ISAC communication and perception integrated data; in low light environment, the visual enhancement mode is turned on, and when the visual sensor fails, it automatically degrades to inertial + active perception combined navigation.
[0056] Implementation Method 4: Lightweight Anti-Poisoning Federated Learning
[0057] The local end compresses the large model into a lightweight model through knowledge distillation, model pruning, and quantization to complete local incremental training; after communication is restored, only the model gradient is uploaded to the cloud; the cloud performs anomaly detection based on the direction consistency or distribution statistical characteristics of the gradient vector, removes malicious gradients, and then uses federated learning to complete global aggregation, generate an updated model, and distribute it to the onboard end to achieve safe iteration.
[0058] Specific Implementation Cases
[0059] Case 1: Smart Return-to-Home Response Due to Limited Urban Communication and GNSS Failure
[0060] When the aircraft enters a building-obstructed area, the communication link is interrupted and the GNSS signal fails. The system determines that communication is restricted according to the flight control cycle, performs a smooth attitude transition to local autonomy, and the first derivative of the power output is continuous. It adopts inertial + vision + radar fusion positioning and automatically activates low light enhancement mode at night. The airflow inversion module optimizes flight energy consumption, dynamically plans obstacle avoidance and return path, and safely returns to the take-off and landing point.
[0061] Case 2: Dynamic table lookup for battery life optimization in low-cost models
[0062] The low-cost drone is equipped with an IMU, pressure sensor, and monocular camera, but no high-end radar; the airflow inversion adopts a dynamic index lookup table method, and the wind resistance is adaptively estimated based on IMU vibration and air pressure changes; communication limitations are determined periodically and switched smoothly to achieve autonomous return and endurance optimization, enabling lightweight and low-cost deployment.
[0063] Case 3: Low Battery Level Emergency Landing
[0064] The aircraft's remaining battery power entered an extremely low level; the downward-facing perception module identified the roads and water below as dangerous areas; the system controlled the aircraft to move horizontally to an open and safe area and perform a smooth emergency landing, with no personnel or property damage and no secondary injuries.
[0065] Case 4: Automatic Degradation and Continued Flight Due to Detection Failure
[0066] The visual sensor failed due to environmental interference; the system automatically downgraded to inertial navigation + active spatial perception mode to maintain stable positioning and real-time obstacle avoidance, and returned autonomously after completing the predetermined task.
[0067] Case 5: Lightweight Anti-Poisoning Model: Self-Evolution
[0068] After the aircraft completes the avoidance maneuver, the onboard unit performs lightweight incremental training; once communication is restored, only the model gradients are uploaded; the cloud detects and removes malicious gradients, completes federated aggregation and global model updates, and achieves safe self-evolution of the model without consuming onboard computing power or leaking original data.
Claims
1. A system for autonomous endurance and intelligent return to home for low-altitude aircraft in communication-constrained environments, characterized in that, It includes a flight control main control unit (1), a flight endurance monitoring unit (2), a multi-source positioning unit (3), an environmental perception unit (4), a distributed communication unit (5), an execution unit (6), and a non-volatile storage unit (7); each unit is electrically connected to the flight control main control unit (1); The system is configured to: perform periodic continuity determination of the communication uplink based on the flight control cycle; enter the local autonomous mode and perform attitude smooth transition when the control loop fails to ensure the continuity of the first derivative of power output; achieve GNSS-free positioning by multi-source sensing fusion; automatically degrade navigation when any sensing unit fails; construct an airflow state observer based on the airframe dynamics and sensor data deviation to optimize global endurance; dynamically plan obstacle avoidance and return path; perform graded emergency response based on remaining endurance; and carry a lightweight model self-evolution module with gradient anomaly detection (71).
2. The system according to claim 1, characterized in that, The multi-source positioning unit (3) includes an inertial navigation module (31), a visual SLAM module (32), an active spatial perception matching module (33), and a geomagnetic sensor; the visual SLAM module (32) supports low-light enhancement and maintains continuous high-precision positioning output when GNSS fails.
3. The system according to claim 1, characterized in that, The environmental perception unit (4) includes an airflow inversion module (41); The airflow state observer inverts the forward airflow vector based on the deviation between the body dynamics model and the airborne sensor data; the drag coefficient can be estimated by using a dynamic index lookup table method when the computing power configuration is low.
4. The system according to claim 1, characterized in that, The flight control cycle is the basic cycle for flight control attitude calculation and power output; communication limitation is determined when no valid uplink command is received for several consecutive control cycles, triggering autonomous switching and starting the attitude smooth transition algorithm.
5. The system according to claim 1, characterized in that, The range monitoring unit (2) is equipped with an energy consumption optimization module to predict the energy consumption of the return path in segments and dynamically adjust the power output to achieve the optimal global energy consumption.
6. The system according to claim 1, characterized in that, The intelligent return path is a dynamically replanned path with real-time obstacle avoidance throughout the process; the environmental perception unit (4) includes a downward perception module (42) for identifying and avoiding dangerous ground areas.
7. The system according to claim 1, characterized in that, The distributed communication unit (5) is equipped with a communication relay module, which automatically scans ground nodes and rebuilds the communication link; the graded emergency response is divided into four levels: normal, low power, ultra-low power, and emergency power, which triggers return or safe landing accordingly.
8. The system according to claim 1, characterized in that, The model self-evolution module (71) uses knowledge distillation to achieve local lightweight incremental training and completes cloud aggregation based on federated learning. Before aggregation, anomaly detection is performed based on the direction consistency or distribution statistical characteristics of gradient vectors to remove malicious gradients. Only the model gradients are uploaded and the original data is not uploaded.
9. A method for autonomous endurance and intelligent return to home in communication-constrained environments for low-altitude aircraft, implemented based on the system described in any one of claims 1-8, characterized in that, include: Monitoring communication link continuity based on flight control cycle; smooth switching to local autonomy and ensuring continuity of the first derivative of power output when control loop fails; multi-source fusion positioning and automatic degradation for sensing failure; optimization of endurance through airflow state observer or dynamic index lookup table; dynamic planning of safe return path; distributed communication relay; emergency response based on power level; safe forced landing; lightweight anti-poisoning model self-evolving.
10. A computer-readable storage medium wherein a program, when executed by a processor, implements the method of claim 9.