An unmanned aerial vehicle real-time semantic perception system and method based on flight state and task adaptation

Abstract

The application discloses a kind of unmanned aerial vehicle real-time semantic perception systems and methods based on flight state and task adaptation.The system includes flight state perception module, task analysis module, adaptive scheduler and configurable perception engine.The core is, according to unmanned aerial vehicle real-time flight state (such as attitude, maneuver mode) and preset task stage (such as search, crossing, landing), dynamically scheduling sensor configuration, selecting algorithm model and adjusting semantic output granularity.The application solves the problem that the resource consumption of existing unmanned aerial vehicle perception system is fixed and the environmental adaptability is poor, realizes the optimal matching of perception ability and task demand, significantly improves the efficiency, safety and overall system robustness of unmanned aerial vehicle autonomous flight in complex dynamic environment.

Description

Technical Field

[0001] This invention belongs to the field of autonomous navigation and intelligent perception technology for unmanned aerial vehicles (UAVs), specifically relating to a system and method that can dynamically adjust the semantic perception model, sensor configuration, and computing resource allocation based on the real-time flight status and preset mission stage of the UAV. Background Technology

[0002] Current semantic understanding solutions for UAVs in dynamic and unstructured environments are mostly general or fixed-mode solutions. Even in simple flight phases (such as high-altitude stable cruise), they still run complex full models, consuming valuable onboard computing and power. The demand for environmental semantic information varies in different mission phases (such as search, close-range observation, and landing), and fixed perception modes cannot provide optimal information tailored to specific needs. During violent maneuvers (such as high-speed turns and evasive maneuvers), image blurring and drastic changes in perspective caused by the UAV's own motion can drastically degrade the performance of general perception models, without any emergency perception mechanism, posing a robustness risk. Furthermore, the perception system and flight control system are relatively independent, requiring complex transformations before being used for control, leading to decision delays and reduced control accuracy in dynamic environments.

[0003] Therefore, existing solutions fail to deeply couple the UAV's flight state (dynamic prior) with the mission phase semantics (mission prior) to guide the perception system in adaptive optimization. Summary of the Invention

[0004] To address these issues, this invention provides a closed-loop "state-task-perception" adaptive framework, transforming the UAV's semantic perception system from a static "data processor" into an "active information acquirer" that is dynamically scheduled based on flight status and mission objectives.

[0005] This application provides a real-time semantic perception system for unmanned aerial vehicles (UAVs) based on flight state and mission adaptation, comprising: a flight state perception module for acquiring internal state data from the UAV flight control system in real time; a mission parsing module for parsing the mission phase semantics of the current flight; an adaptive scheduler, communicatively connected to both the flight state perception module and the mission parsing module, for generating dynamic perception configuration instructions based on the internal state data and the mission phase semantics, according to a predefined mapping strategy; a configurable perception engine, communicatively connected to the adaptive scheduler, for receiving the dynamic perception configuration instructions and dynamically adjusting its operating parameters accordingly to perform environmental perception and output environmental semantic information; and a flight control and decision-making module for receiving the environmental semantic information and generating UAV control instructions.

[0006] According to the real-time semantic perception system for unmanned aerial vehicles (UAVs) based on flight state and mission adaptation provided by the present invention, the internal state data includes at least one of the following: attitude angle and angular velocity data provided by the inertial measurement unit, current control mode and current control command provided by the flight controller, motor speed data, and relative altitude or altitude change rate data provided by the barometer or ranging sensor.

[0007] According to the real-time semantic perception system for unmanned aerial vehicles (UAVs) based on flight state and mission adaptation provided by the present invention, the mission phase semantics include at least one of the following: area search, target tracking, narrow space traversal, precise landing, and emergency return to base.

[0008] According to the real-time semantic perception system for unmanned aerial vehicles based on flight state and mission adaptation provided by the present invention, the configurable perception engine includes a dynamically configurable sensor subsystem and a data processing subsystem; the dynamic perception configuration instructions include sensor scheduling instructions for the sensor subsystem and algorithm scheduling instructions for the data processing subsystem.

[0009] According to the real-time semantic perception system for unmanned aerial vehicles based on flight state and mission adaptation provided by the present invention, the sensor scheduling command is used to control at least one of the following: enabling or disabling a specific type of sensor; adjusting the sampling frequency or resolution of the sensor; and switching the working mode of the sensor.

[0010] According to the real-time semantic perception system for unmanned aerial vehicles based on flight state and mission adaptation provided by the present invention, the algorithm scheduling instruction is used to control at least one of the following: loading a specified neural network model from a pre-stored model library; setting the image region or point cloud region of interest for data processing; adjusting the execution frequency of the semantic perception algorithm; and setting the output granularity of the semantic information related to the environment.

[0011] According to the real-time semantic perception system for unmanned aerial vehicles (UAVs) based on flight state and mission adaptation provided by the present invention, the adaptive scheduler pre-stores a mapping strategy table, which defines the correspondence between the internal state data, the mission phase semantics, and the dynamic perception configuration instructions; wherein, the correspondence includes: when the internal state data indicates that the UAV is in an uncontrolled rotation state with an angular velocity exceeding a predetermined threshold, the adaptive scheduler generates an instruction for switching the configurable perception engine to an emergency response mode based on an event camera or pure inertial-assisted optical flow.

[0012] According to the real-time semantic perception system for unmanned aerial vehicles based on flight state and mission adaptation provided by the present invention, the adaptive scheduler is further configured to: select a semantic output type that matches the mission objective based on the semantics of the mission phase; wherein, in the precision landing phase, the semantic output type is a high-precision terrain descent grid map; and in the target tracking phase, the semantic output type is a fine pixel-level mask of the target.

[0013] This invention also provides a real-time semantic perception method for unmanned aerial vehicles (UAVs) based on flight state and mission adaptation. The method includes: acquiring internal state data of the UAV in real time; parsing the semantics of the current flight mission phase; generating dynamic perception configuration instructions by querying a predefined mapping strategy based on the internal state data and the mission phase semantics; dynamically adjusting the operating parameters of the perception system according to the dynamic perception configuration instructions; performing environmental perception using the adjusted perception system and outputting environmental semantic information.

[0014] According to the real-time semantic perception method for UAVs based on flight state and mission adaptation provided by the present invention, the step of generating dynamic perception configuration instructions based on internal state data and mission stage semantics includes: when the mission stage semantics is narrow forest crossing and the internal state data shows drastic attitude changes, generating instructions to enable omnidirectional vision sensors, load dynamic real-time localization and mapping models, increase perception frequency and output a three-dimensional densely occupied grid map.

[0015] This invention provides a real-time human-machine semantic perception system and method based on flight state and task adaptation. Driven by both flight state and task phase, it dynamically configures perception models and sensor resources, precisely matching computational load, energy consumption, and real-time requirements. This solves the resource waste problem of fixed perception modes and significantly improves system efficiency. For states such as violent UAV maneuvers, it adaptively switches perception strategies (e.g., activating event cameras, switching lightweight models), effectively overcoming the performance drop caused by image blurring and drastic viewpoint changes in traditional algorithms. This enhances the robustness of semantic understanding in complex dynamic environments and ensures reliable perception under extreme conditions. The perception output granularity is strongly correlated with the task (e.g., outputting a landability grid during landing, and a pixel-level mask during tracking), directly providing the most operational semantic understanding results for flight control and decision-making. It provides high-value, task-oriented semantic information, improving the intelligence and accuracy of overall autonomous behavior. Deeply coupling the internal state of flight control to the perception loop achieves a paradigm shift from "passive processing" to "active prediction and adaptation," forming a close synergy between perception and control, building system-level synergistic advantages, and systematically improving the UAV's survivability and task completion capabilities in unstructured environments. Attached Figure Description

[0017] Other objects and results of the invention will become more apparent and readily understood with reference to the following description taken in conjunction with the accompanying drawings and the contents of the claims, and with a more complete understanding of the invention. In the drawings: Figure 1 This is a schematic diagram of a real-time semantic perception system for unmanned aerial vehicles (UAVs) based on flight state and mission adaptation, provided by the present invention. Figure 2 This is a real-time semantic perception system for unmanned aerial vehicles (UAVs) based on flight state and mission adaptation, as provided in Embodiment 1 of the present invention. Figure 3 This is a real-time semantic perception system for unmanned aerial vehicles (UAVs) based on flight status and mission adaptation, provided according to Embodiment 2 of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] The core of this invention lies in constructing an intelligent framework capable of dynamically adjusting the configuration of a UAV's semantic perception system based on its flight status and mission phase. This system achieves optimal allocation of perception resources and maximizes mission efficiency through deep coupling of flight control internal data and mission context.

[0021] like Figure 1As shown, the UAV real-time semantic perception system based on flight state and mission adaptation provided by this invention is an intelligent software-hardware collaborative platform deeply embedded in the original flight control architecture of the UAV. Logically, the system consists of a closed loop of five core functional modules, interconnected via a high-speed internal bus and data interface, forming a complete closed loop from environmental information acquisition and intelligent analysis to flight decision-making. The flight state perception module 101 directly connects to the flight control bus, continuously acquiring UAV attitude, angular velocity, control commands, and motor speed, among other motion data. The mission parsing module 103 is responsible for translating high-level mission commands into specific stage semantics such as "tower inspection" or "window delivery" in real time. The adaptive scheduler 105, as the core hub, makes millisecond-level decisions on the aforementioned state and mission information based on a built-in mapping strategy library, generating dynamic configuration commands. After receiving commands, the configurable perception engine 107 can schedule sensors in real time (e.g., activate fisheye lenses or lidar) and switch algorithm models (e.g., load a dedicated recognition network) to ensure that the perception capability accurately matches the current requirements. Ultimately, the optimized semantic information (such as the landing area rating map) is sent to the flight control and decision-making module 109 to drive the UAV's actions, while the newly generated flight status is fed back in real time, forming a continuously self-optimizing intelligent closed loop.

[0022] Specifically, the system's data flow originates from the flight status perception module 101. This module does not refer to the perception of the external environment, but rather specifically to the precise monitoring and acquisition of the UAV's internal operating status. It directly connects to the flight control computer's underlying data bus, reading and preprocessing raw attitude angle and three-axis angular velocity data from the inertial measurement unit at a frequency of no less than 100Hz, current control mode commands and control surface parameters from the flight control kernel, real-time speed and current data of each brushless motor, and relative altitude information from the barometer or ultrasonic sensor. The core task of this module is to provide the upper-level system with a precise, low-latency digital twin report on "how the UAV is moving."

[0023] Working in parallel with the state perception module is the task parsing module 103. This module acts as an intelligent translator between the task planning system and the underlying perception-control loop, receiving input from high-level commands from the ground control station or the onboard mission computer. It can parse and decompose abstract tasks such as "perform inspection of power pole number 3" or "deliver goods to window number 3 on the 5th floor of Building B" in real time into a series of stages with clear semantics and boundaries, such as "area search," "target tracking," "narrow passage crossing," "precision landing," and "emergency return." Each stage defines the core objectives and behavioral paradigms of the UAV in that phase, providing fundamental contextual basis for subsequent perception resource allocation.

[0024] The "intelligent scheduling hub" of the entire system is the adaptive scheduler 105. This module receives real-time data streams from the two modules mentioned above and internally maintains a structured "state-task-configuration" mapping strategy table. This table is not a simple lookup table, but a rule engine containing logical judgments and priority arbitration. For example, when it simultaneously receives an emergency state signal of "angular velocity exceeding the threshold" and a task instruction of "detailed shooting," its built-in rules will prioritize responding to the state signal and forcibly trigger the emergency perception mode. Based on real-time input, the scheduler makes decisions within milliseconds and outputs a set of structured, executable, dynamically configured instructions to the next module.

[0025] It should be noted that the adaptive scheduler 105 has preset intelligent mapping rules between state / task and perception configuration, as detailed in the table below: Flight status / mission phase Typical characteristics Adaptive Perception Configuration Strategy High-speed cruise Stable attitude and high forward speed It employs a forward-looking narrow field-of-view camera and a lightweight detection model to focus on distant obstacles; reduces the sensing frequency to save power; and outputs coarse-grained obstacle information. Narrow forest crossing High angular velocity, drastic attitude changes, and numerous nearby obstacles. Enable omnidirectional fisheye camera / event camera; load a lightweight dynamic SLAM model with anti-blurring; increase perception frequency; output a dense 3D occupancy grid map. Hovering Target Observation With a slight shift in posture, the target came into view. The system schedules the high-definition camera on the gimbal; loads a high-precision target recognition and segmentation model; locks the target region in the perception area; and outputs a fine pixel-level mask and category of the target. Precision landing phase The vertical speed is obvious, and the height decreases. Enhance the downward-looking sensors (LiDAR / downward-looking camera); enable a dedicated landing area assessment model; output high-precision terrain slope and flatness semantic maps. Uncontrolled rotation (emergency response) Angular velocity exceeds threshold (IMU directly triggered) Immediately switch to event camera stream or pure IMU-assisted sparse optical flow obstacle avoidance algorithm to bypass complex semantic understanding, achieve millisecond-level reactive obstacle avoidance, and prioritize survival. High-speed cruise Stable attitude and high forward speed It employs a forward-looking narrow field-of-view camera and a lightweight detection model to focus on distant obstacles; reduces the sensing frequency to save power; and outputs coarse-grained obstacle information.

[0026] It should be noted that the adaptive scheduler 105 is also used to: select a semantic output type that matches the task objective according to the task phase semantics 104; wherein, in the precision landing phase, the semantic output type is a high-precision terrain descent raster map; and in the target tracking phase, the semantic output type is a fine pixel-level mask of the target.

[0027] The receiver and executor of these configuration instructions is the configurable perception engine 107, which is the system's "agile execution limb." Physically, this engine comprises heterogeneous sensor arrays (such as visible light cameras, LiDAR, and event cameras) and heterogeneous computing units (such as GPUs and DSPs); logically, it is a software container that supports dynamic loading. Based on the scheduler's instructions, the following operations can be dynamically performed: at the sensor level, enabling or disabling specific types of sensors (such as the main camera, fisheye camera, and LiDAR), adjusting the sampling frequency or resolution of the sensors, or switching the operating mode of the sensors; at the algorithm level, quickly loading or unloading specified neural network model weight files from the pre-stored model library, switching the core visual odometry or SLAM algorithm flow, setting the image region or point cloud region of interest for data processing (e.g., processing only the central image region for long-distance tracking, or processing edge regions for omnidirectional obstacle avoidance), and adjusting the execution frequency of the semantic perception algorithm; at the output level, controlling the output granularity of environmental semantic information (e.g., outputting a dense "landability" grid during landing, and only outputting "obstacle" bounding boxes during cruise). This allows the same set of physical hardware to instantly switch between two completely different perception modes: "long-distance target search" and "near-distance obstacle avoidance and traversal".

[0028] The entire loop ultimately converges into the flight control and decision-making module 109. This module receives a semantic information stream from the configurable perception engine, filtered and optimized by the task and state context. For example, during the landing phase, it receives not raw image frames or cluttered point clouds, but a directly usable landing area rating map labeled with "flatness" and "slope". Based on this high-quality, highly relevant environmental perception, combined with higher-level mission objectives, the module performs path planning, obstacle avoidance decisions, and ultimately generates specific motor control commands to drive the UAV to perform actions. These newly generated control commands and the resulting changes in the aircraft's state are immediately captured by the flight state perception module, thus initiating the next cycle of adaptive adjustment, forming a continuously optimizing, self-adjusting intelligent perception-control closed loop.

[0029] Through the precise coordination of the above five modules, this system achieves a fundamental transformation from a rigid, fixed perception pipeline to a flexible, context-driven perception agent.

[0030] like Figure 2 As shown, the present invention also provides an embodiment: During power line inspection, when the task parsing module 102 determines that the "tower-around inspection" stage has been entered, and the state perception module 101 detects that the UAV is entering a predetermined orbiting mode (such as a preset orbiting radius and angular velocity), the adaptive scheduler (103) starts a dedicated mapping strategy. At this time, sensor scheduling: omnidirectional fisheye camera (201) is enabled to cover the omnidirectional field of view of the tower and surrounding lines; forward solid-state lidar (202) is started simultaneously to focus on the accurate ranging of the tower structure. Algorithm scheduling: lightweight component recognition models for power equipment (such as insulator and vibration damper recognition) are loaded from the model library, and a tightly coupled vision-inertial-lidar SLAM algorithm is started to cope with the challenges of monotonous background texture and complex structure during close-range orbiting. Output scheduling: semantic output focuses on the three-dimensional sparse semantic point cloud of the tower and conductor, and highlights the specific components and their states (such as "insulator - normal"). Meanwhile, the sensing frequency is increased to 30Hz to ensure real-time updates of obstacle information (such as bird nests and foreign objects) on the tower surface during rapid circling. Compared with traditional fixed inspections using high-definition gimbal cameras, this embodiment, through state and task adaptive sensing configuration, achieves real-time obstacle perception without blind spots and automatic identification and positioning of key components during circling, with the same computing power, thus simultaneously improving inspection efficiency and safety.

[0031] like Figure 3As shown, the present invention also provides an embodiment two: During logistics delivery, when the UAV receives the "deliver through window" instruction, the task parsing module (102) decomposes it into three sub-stages: "approach recognition", "hovering adjustment" and "crossing / delivery". The adaptive scheduler (103) dynamically adjusts according to each sub-stage. "Approach recognition" stage: The UAV flies towards the target building. The scheduler instructs the perception engine to load a general building and window detection model and uses binocular vision to generate a rough depth map to locate the approximate area of ​​the target window. "Hovering adjustment" stage: The UAV hovers outside the window. The state perception module (101) detects that the UAV has entered a stable hovering mode. At this time, the scheduler immediately switches strategies: enables high-resolution downward-looking lidar (203) and zoom optical camera (204); loads a fine "window frame and indoor passable space segmentation model"; the algorithm focuses on analyzing the window opening status, window frame size and indoor ground landing area. "Crossing / delivery" stage: The crossing action begins. The state perception module (101) detects that the UAV generates a precise axial translation speed. The scheduler switches to a high-speed (e.g., 60Hz) sparse visual odometry and a near-field infrared obstacle avoidance sensor, outputting a simplified "passable" and "impassable" binary spatial grid to ensure agility and safety during passage. Compared to traditional logistics drones that rely solely on GPS and preset waypoints, this embodiment solves the core challenges of dynamic recognition of non-standard windows, seamless integration of indoor and outdoor environments without GPS, and centimeter-level precise passage through multi-stage adaptive perception, achieving automated and accurate delivery in complex urban scenarios.

[0032] This invention also provides a real-time semantic perception method for unmanned aerial vehicles (UAVs) based on flight state and mission adaptation, the method comprising: Real-time acquisition of the drone's internal status data 102; Semantics of the current flight mission phase 104; Based on the internal state data 102 and the task stage semantics 104, a dynamic perception configuration instruction 106 is generated by querying a predefined mapping strategy. According to the dynamic sensing configuration instruction 106, the operating parameters of the sensing system are dynamically adjusted. The adjusted perception system is used to perform environmental perception and outputs semantic information about the environment 108.

[0033] The step of generating dynamic-aware configuration instructions (106) based on internal state data (102) and task stage semantics (104) includes: When the task phase semantics (104) is narrow forest crossing and the internal state data (102) shows drastic attitude changes, instructions are generated to enable omnidirectional vision sensors, load dynamic real-time localization and mapping models, increase perception frequency, and output a three-dimensional densely occupied grid map.

[0034] This invention creatively uses the internal state data of the flight control system as a priori input for scheduling the perception system, a feature not found in ground robots. With limited onboard resources, intelligent scheduling achieves optimal matching between perception capabilities and task requirements, fundamentally improving efficiency and practicality. For extreme situations such as violent maneuvers by the UAV, a perception mode degradation mechanism from "deep thinking" to "response" is designed, significantly enhancing the system's survivability in dynamic environments. This invention is not an improvement on a single algorithm, but an innovation in system architecture and scheduling methods, offering broader protection and easy integration with existing perception algorithms. The system and method described in this invention can be widely applied in fields requiring advanced autonomous operation in highly dynamic, unstructured environments, such as UAV power line inspection, logistics delivery, emergency rescue, and agricultural plant protection. It effectively improves the level of operational intelligence, safety, and reliability, demonstrating significant industrial applicability.

[0035] The real-time semantic perception system and method for unmanned aerial vehicles (UAVs) based on flight state and task adaptation according to the present invention have been described above by way of example with reference to the accompanying drawings. However, those skilled in the art should understand that various modifications can be made to the real-time semantic perception system and method for UAVs based on flight state and task adaptation proposed in the present invention without departing from the scope of the invention. Therefore, the scope of protection of the present invention should be determined by the content of the appended claims.

Claims

1. A real-time semantic perception system for unmanned aerial vehicles (UAVs) based on flight state and mission adaptation, characterized in that, include: The flight status perception module (101) is used to acquire internal status data (102) from the UAV flight control system in real time. The mission parsing module (103) is used to parse the semantics of the mission phase of the current flight (104). An adaptive scheduler (105) is communicatively connected to the flight state perception module (101) and the task parsing module (103), respectively, and is used to generate dynamic perception configuration instructions (106) according to the internal state data (102) and the task stage semantics (104) based on a predefined mapping strategy. A configurable perception engine (107), communicatively connected to the adaptive scheduler (105), is used to receive the dynamic perception configuration instruction (106) and dynamically adjust its operating parameters accordingly to perform environmental perception and output environmental semantic information (108); and, The flight control and decision module (109) is used to receive the environmental semantic information (108) and generate UAV control commands (110).

2. The system according to claim 1, characterized in that, The internal state data (102) includes at least one of the following: attitude angle (1021) and angular velocity (1022) provided by the inertial measurement unit, current control mode (1023) and current control command (1024) provided by the flight controller, motor speed (1025), and relative altitude (1026) or altitude change rate (1027) provided by the barometer or distance sensor.

3. The system according to claim 1 or 2, characterized in that, The task phase semantics (104) include at least one of the following: area search (1041), target tracking (1042), narrow space crossing (1043), precision landing (1044), and emergency return (1045).

4. The system according to claim 1, characterized in that, The configurable sensing engine (107) includes a dynamically configurable sensor subsystem (1071) and a data processing subsystem (1072). The dynamic sensing configuration instruction (106) includes a sensor scheduling instruction (1073) for the sensor subsystem (1071) and an algorithm scheduling instruction (1074) for the data processing subsystem (1072).

5. The system according to claim 4, characterized in that, The sensor scheduling instruction (1073) is used to control at least one of the following: Enable or disable specific types of sensors; Adjust the sampling frequency or resolution of the sensor; Switch the operating mode of the sensor.

6. The system according to claim 4, characterized in that, The algorithm scheduling instruction (1074) is used to control at least one of the following: Quickly load or unload specified neural network model weight files from a pre-stored model library; Define the image region or point cloud region of interest for data processing; Adjust the execution frequency of the semantic awareness algorithm; Set the output granularity of the aforementioned environmental semantic information.

7. The system according to claim 1, characterized in that, The adaptive scheduler (105) has a pre-stored mapping strategy table, which defines the correspondence between the internal state data (102), the task stage semantics (104), and the dynamic perception configuration instructions (106). The correspondence includes: when the internal state data indicates that the UAV is in an uncontrolled rotation state with an angular velocity exceeding a predetermined threshold, the adaptive scheduler generates an instruction to switch the configurable perception engine (107) to an emergency perception mode based on an event camera or pure inertial-assisted optical flow.

8. The system according to claim 1, characterized in that, The adaptive scheduler (105) is also used for: Based on the semantics of the task phase (104), a semantic output type matching the task objective is selected; wherein, in the precision landing phase, the semantic output type is a high-precision terrain descent raster map; and in the target tracking phase, the semantic output type is a fine pixel-level mask of the target.

9. A real-time semantic perception method for unmanned aerial vehicles (UAVs) based on flight state and mission adaptation, characterized in that, The method, applied to the system as described in any one of claims 1-8, comprises: Real-time acquisition of internal status data of the drone (102); Parse the semantics of the current flight mission phase (104); Based on the internal state data (102) and the task stage semantics (104), a dynamic perception configuration instruction (106) is generated by querying a predefined mapping strategy. According to the dynamic sensing configuration instruction (106), the operating parameters of the sensing system are dynamically adjusted. The adjusted perception system is used to perform environmental perception and output semantic information about the environment (108).

10. The method according to claim 9, characterized in that, The generation of dynamically aware configuration instructions (106) based on internal state data (102) and task stage semantics (104) includes: When the task phase semantics (104) is narrow forest crossing and the internal state data (102) shows drastic attitude changes, instructions are generated to enable omnidirectional vision sensors, load dynamic real-time localization and mapping models, increase perception frequency, and output a three-dimensional densely occupied grid map.

Images