A method, system, device and medium for adaptive rolling trajectory planning of a UAV

Abstract

This invention discloses an adaptive rolling trajectory planning method, system, device, and medium for unmanned aerial vehicles (UAVs), relating to the field of UAV trajectory planning technology. The method obtains constrained flow field disturbances through computational fluid dynamics and closed-loop disturbance rejection simulation, generating an aerodynamic safety boundary hard constraint model based on the dynamic changes in UAV velocity and attitude. It utilizes multi-source sensor point cloud fusion to obtain environmental geometric features such as channel width, generating a constrained environment risk state representation vector. This risk state representation vector is input into a reinforcement learning parameter tuning model to smoothly update the target weight parameters for trajectory optimization. The aerodynamic boundary hard constraint model and the updated objective function are jointly optimized to obtain a reference trajectory. Closed-loop executability verification is performed on the reference trajectory based on the safety boundary margin. If an aerodynamic failure risk exists, a safety avoidance strategy is triggered, effectively solving the problem of uncontrolled UAV crashes in confined spaces due to neglecting aerodynamic boundaries, thus improving flight safety.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) trajectory planning technology, and in particular to an adaptive rolling trajectory planning method, system, device and medium for UAVs that integrates aerodynamic safety boundary constraints. Background Technology

[0002] In recent years, the demand for drones in disaster relief, confined space inspection, underground space exploration, and complex building interior operations has increased significantly. These application environments are typically characterized by limited space, dense obstacles, and irregular structures. In such complex environments, drones often need to perform significant attitude adjustments and drastic speed changes when navigating and avoiding obstacles. Existing drone trajectory planning methods mainly include sampling-based path planning, graph search-based path planning, and model predictive control (MPC)-based optimization planning. These methods typically achieve geometric obstacle avoidance and basic motion constraints on the planned trajectory by setting obstacle distance constraints, collision penalty terms, or simplified mass kinematics and dynamics conditions.

[0003] However, existing technologies have the following key limitations: First, safety constraints primarily focus on geometric obstacle avoidance, lacking systematic quantitative modeling of the aerodynamic safety flight boundaries of UAVs. In real flight mechanisms, the state variables of UAVs exhibit highly coupled aerodynamic limits. When a UAV makes sharp turns or traverses at high speeds in narrow passages, the intense aerodynamic effects drastically compress the system's stability margin. If the goal is merely to "avoid hitting walls geometrically," when the flight state approaches the aerodynamic limits, it can easily lead to motor command saturation in the underlying controller, attitude instability, or even a crash.

[0004] Secondly, existing technologies do not uniformly consider aerodynamic safety boundaries as mandatory hard constraints in rolling time-domain optimization. This results in the generated reference trajectories often being feasible only in a kinematic or simplified dynamic sense, and potentially approaching or exceeding the system's safe flight limits during actual execution, leading to a serious disconnect between upper-level planning and lower-level control.

[0005] Finally, the weight parameters in the planning objective function are usually fixed, lacking a mechanism for adaptive dynamic adjustment based on environmental risk status. Most methods employ static and conservative obstacle avoidance boundaries, making it difficult to balance intrinsic safety and flight efficiency requirements under different confined environments (such as passages that suddenly narrow or widen). Summary of the Invention

[0006] To address the technical challenge of existing UAV trajectory planning methods struggling to balance dynamic aerodynamic safety boundaries with environmentally adaptive planning in complex and constrained environments, this invention proposes a novel UAV adaptive rolling trajectory planning method, system, device, and medium that integrates aerodynamic safety boundary constraints. This method deeply integrates aerodynamic safety flight boundary constraints with a rolling time-domain trajectory optimization framework, enabling adaptive online adjustment of planning target parameters and comprehensively improving the practical executability of the trajectory and the inherent safety of the flight system.

[0007] In a first aspect, the present invention provides an adaptive rolling trajectory planning method for unmanned aerial vehicles (UAVs), comprising the following steps: The three-dimensional geometric model and initial aerodynamic performance indicators of the UAV flight operation platform are obtained, a system closed-loop control simulation model is constructed, flow field disturbances are obtained through computational fluid dynamics and input into the system closed-loop control simulation model, and a dynamic aerodynamic safety boundary hard constraint model of the UAV is generated and output. Acquire multi-source point cloud observation data collected by airborne heterogeneous sensors, transform the multi-source point cloud observation data to the local body coordinate system for fusion and denoising, extract the geometric features of the confined environment, and generate and output the confined environment risk state characterization vector. A rolling time-domain trajectory optimization objective function including planning target weight parameters is constructed. The constrained environment risk state representation vector is used as the state space input reinforcement learning adaptive parameter tuning model to calculate and output the smoothly updated planning target weight parameters. The dynamic aerodynamic safety boundary hard constraint model and the geometric features of the confined environment are used together as constraints for optimization, and the rolling time-domain trajectory optimization objective function is updated using the smoothly updated planning objective weight parameters to solve and output the reference trajectory. The reference trajectory is subjected to closed-loop executability verification. If the preset safety boundary margin is met, the reference trajectory is output to the underlying flight control system for execution. If it is not met, a safety action avoidance strategy is triggered and an avoidance control command is output.

[0008] As an optional implementation of the first aspect of this application, generating and outputting a dynamic aerodynamic safety boundary hard constraint model for the UAV specifically includes: based on the initial aerodynamic performance indicators, meshing the three-dimensional geometric model of the UAV flight operation platform; calculating transient values ​​under different constrained environments and at different approximation distances using a sliding mesh calculation method; extracting the fluctuation range and peak parameters of the platform's aerodynamic forces and moments as real-time external disturbances; inputting the real-time external disturbances into the system closed-loop control simulation model to obtain time-domain response performance indicators and frequency-domain response performance indicators under disturbance conditions, wherein the time-domain response performance indicators include maximum overshoot and settling time, and the frequency-domain response performance indicators include amplitude margin and phase margin; and constructing a safety boundary hard constraint model. The system can determine the safety performance judgment set, which is set as follows: the maximum overshoot of the closed-loop system is less than or equal to the maximum allowable overshoot, the settling time is less than or equal to the longest allowable settling time, the amplitude margin is greater than or equal to the minimum amplitude margin, and the phase margin is greater than or equal to the minimum phase margin. Under the set flight state, the system gradually approaches the physical boundary of the environment in each direction of the body coordinate system, and records the critical approach distance when the comprehensive performance evaluation index first falls out of the safety performance judgment set. By traversing the set total thrust and the speed and attitude ranges under the confined space, the critical state of the UAV flight operation platform when approaching the constrained environment is quantified, and a three-dimensional spatial envelope surface function that dynamically changes with the platform's motion speed and attitude is fitted and constructed, and the output is used as the hard constraint model of the UAV's dynamic aerodynamic safety boundary.

[0009] As an optional implementation of the first aspect of this application, the dynamic aerodynamic safety boundary hard constraint model transforms its quantized three-dimensional spatial envelope surface into a mathematical analytical expression embedded in the rolling time domain trajectory optimization, forming an explicit hard constraint condition. The explicit hard constraint condition is that, within the rolling prediction time domain, the predicted distance between the UAV's future predicted spatial position and the perceived position of environmental obstacles not only satisfies the static geometric collision avoidance distance, but the predicted distance is also greater than or equal to the dynamic aerodynamic safety boundary thickness calculated by substituting the current predicted motion speed and predicted attitude into the three-dimensional spatial envelope surface function.

[0010] As an optional implementation of the first aspect of this application, multi-source point cloud observation data collected by airborne heterogeneous sensors is acquired, and the multi-source point cloud observation data is converted to a local body coordinate system for fusion and denoising to extract the geometric features of the restricted environment. Specifically, this includes: establishing a local body coordinate system with the current centroid of the UAV as the origin, the longitudinal axis of the body as the longitudinal axis, the vertical upward as the vertical axis, and the horizontal axis determined by the right-hand rule; acquiring the external parameter calibration matrix of each of the airborne heterogeneous sensors, the external parameter calibration matrix including a rotation matrix and a translation vector, and uniformly converting the multi-source point cloud observation data of the airborne heterogeneous sensors in their respective coordinate systems to the local body coordinate system through the external parameter calibration matrix; aligning and fusing the converted multi-source point cloud observation data, performing data simplification and environmental denoising based on spatial point density features to extract the local effective obstacle point cloud set at the current moment; and extracting the restricted environment geometric features along the predicted flight trajectory direction of the UAV in the local effective obstacle point cloud set, the restricted environment geometric features including the distance to the nearest obstacle in front, the passable lateral width in front, and the passable vertical height in front.

[0011] As an optional implementation of the first aspect of this application, generating and outputting a restricted environment risk state representation vector specifically includes: obtaining the sampling period within a discrete time interval; calculating the difference between the current time and the previous time interval for the passable horizontal width using backward difference and dividing it by the sampling period to obtain the real-time change rate of the passable horizontal width; calculating the difference between the current time and the previous time interval for the passable vertical height using backward difference and dividing it by the sampling period to obtain the real-time change rate of the passable vertical height; and combining the current time for the passable horizontal width, the real-time change rate of the passable horizontal width, the passable vertical height, and the real-time change rate of the passable vertical height in a predetermined order to construct, generate, and output the restricted environment risk state representation vector.

[0012] As an optional implementation of the first aspect of this application, calculating and outputting the smoothly updated planning target weight parameters specifically includes: setting the rolling time-domain trajectory optimization objective function to include a cost term representing flight efficiency and trajectory tracking error, a cost term representing control command smoothness, and a penalty term representing obstacle avoidance safety margin, and configuring corresponding initial planning target weight parameters for each cost term and penalty term; inputting the generated restricted environment risk state representation vector as the state space feature at the current moment into the reinforcement learning adaptive parameter tuning model, the reinforcement learning adaptive parameter tuning model performing network forward propagation calculation based on the channel width and approximation velocity change characteristics of the current restricted environment, and outputting the target action values ​​of each planning target weight parameter; using a preset upper and lower limit function to truncate and restrict each target action value within a preset physical operation range to obtain the limited weight parameters; based on a set smoothing coefficient, performing a weighted summation calculation of the planning target weight parameters at the previous control moment and the limited weight parameters at the current moment, calculating and outputting the final smoothly updated planning target weight parameters.

[0013] As an optional implementation of the first aspect of this application, the closed-loop executability verification of the reference trajectory specifically includes: extracting the initial spatial state point on the generated reference trajectory, and obtaining the actual lateral distance of the obstacle at the current moment using airborne sensor data; extracting the corresponding dynamic aerodynamic safety boundary thickness based on the current flight state, subtracting the dynamic aerodynamic safety boundary thickness from the actual obstacle lateral distance, and calculating the safety boundary margin at the current moment; determining whether the optimization solver has successfully found a feasible solution and whether the safety boundary margin is greater than zero; if the solver has successfully found a feasible solution and the safety boundary margin is greater than zero, the verification is passed, and the reference trajectory is output to the underlying flight control system for execution; if the objective function of the rolling time-domain trajectory optimization has no solution, or the safety boundary margin is less than or equal to zero, a safety action avoidance strategy is triggered, the planner takes over the flight control commands and outputs commands including hovering, emergency braking, or retreating along the original safety trajectory to the actuator.

[0014] Secondly, embodiments of this application provide an adaptive rolling trajectory planning system for unmanned aerial vehicles (UAVs), comprising: The dynamic aerodynamic safety boundary hard constraint modeling module is used to obtain the three-dimensional geometric model and initial aerodynamic performance indicators of the UAV flight operation platform, construct the system closed-loop control simulation model, obtain the flow field disturbance through computational fluid dynamics and input it into the system closed-loop control simulation model, and generate and output the UAV dynamic aerodynamic safety boundary hard constraint model. The confined environment perception and risk characterization module is used to acquire multi-source point cloud observation data collected by airborne heterogeneous sensors, convert the multi-source point cloud observation data to the local body coordinate system for fusion and noise reduction, extract the geometric features of the confined environment, and generate and output the confined environment risk state characterization vector. The reinforcement learning adaptive planning parameter tuning module is used to construct a rolling time-domain trajectory optimization objective function including planning target weight parameters, and to input the restricted environment risk state representation vector as the state space into the reinforcement learning adaptive parameter tuning model, and to calculate and output the smoothly updated planning target weight parameters. The rolling trajectory optimization solution and safety verification execution module is used to take the dynamic aerodynamic safety boundary hard constraint model and the geometric features of the confined environment as constraints for optimization, and update the rolling time-domain trajectory optimization objective function using the smoothly updated planning objective weight parameters to solve and output the reference trajectory; perform closed-loop executability verification on the reference trajectory, and if the preset safety boundary margin is met, output the reference trajectory to the underlying flight control system for execution; if not, trigger the safety action avoidance strategy and output avoidance control commands.

[0015] Thirdly, embodiments of this application provide an electronic device, which includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the steps of the method described in the first aspect.

[0016] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention, for the first time, constructs a dynamic aerodynamic safety flight boundary model for UAVs through offline flow field mechanism analysis and closed-loop disturbance rejection capability analysis, and embeds it as an explicit hard constraint into the rolling time-domain trajectory planning (MPC) process. Compared with existing methods that only consider static geometric distance collision avoidance, this scheme ensures that the generated trajectory strictly conforms to the real aerodynamic capability limits that the underlying controller can stably execute (such as the ability to suppress downwash airflow wall effect rebound), completely avoiding the pseudo-safe trajectory problem of "no collision in space, but loss of control and crash due to motor aerodynamic saturation in physical execution," and significantly improving the actual executability of the trajectory.

[0018] 2. This invention innovatively introduces a reinforcement learning adaptive parameter tuning mechanism based on environmental risk state representation (such as the differential trend of narrowing and widening of restricted passages). It overcomes the shortcomings of fixed and rigid weights in traditional optimization objective functions, enabling automatic enhancement of safety and smoothness constraints to suppress aggressive UAV maneuvers in high-risk, narrow environments, and automatic improvement of efficiency weights in low-risk, spacious environments. This achieves the optimal dynamic balance between safety and flight efficiency in variable, unknown, and restricted environments.

[0019] 3. This invention constructs a "closed-loop executability verification and safety backoff mechanism" with safety boundary margin monitoring. When the optimization solver has no solution or the aerodynamic safety margin is exhausted, it can automatically trigger actions such as emergency braking or safety retreat to perform bottom-line defense. This mechanism endows the flight system with excellent disturbance resistance and absolute flight safety assurance capabilities when operating in extremely dense obstacles and narrow underground spaces. Attached Figure Description

[0020] Figure 1 A flowchart of an adaptive rolling trajectory planning method for unmanned aerial vehicles provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the downwash airflow disturbance when an unmanned aerial vehicle (UAV) approaches the ground at low speed, provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of airflow deformation when a drone flies at high speed and approaches the ground, provided in an embodiment of the present invention. Figure 4 A schematic diagram of asymmetric eddies and torque disturbances generated when a UAV approaches a boundary in an tilted attitude, as provided in an embodiment of the present invention; Figure 5 This is a schematic diagram comparing dynamic aerodynamic safety boundaries at different motion speeds provided in an embodiment of the present invention; Figure 6 A schematic diagram of the asymmetric aerodynamic safety boundary of an unmanned aerial vehicle in a tilted attitude provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an adaptive rolling trajectory planning system for unmanned aerial vehicles (UAVs) provided in the second embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0023] Example 1 This invention proposes an adaptive rolling trajectory planning method for unmanned aerial vehicles (UAVs) that integrates aerodynamic safety boundary constraints. This method designs a complete closed-loop replanning process encompassing "environmental perception—risk characterization—parameter update—optimization solution—execution feedback." For example... Figure 1 As shown, the core mechanism and specific implementation details of this invention are explained in detail below, step by step: S1: Obtain the three-dimensional geometric model and initial aerodynamic performance indicators of the UAV flight operation platform, construct the system closed-loop control simulation model, obtain the flow field disturbance through computational fluid dynamics and input it into the system closed-loop control simulation model, and generate and output the UAV dynamic aerodynamic safety boundary hard constraint model.

[0024] In real, confined passages, the downwash airflow generated by the high-speed rotating propellers of multi-rotor UAVs creates strong rebound vortices and aerodynamic coupling disturbances (i.e., ground effect, wall effect, etc.) when it encounters walls, ceilings, or ruin surfaces, drastically compressing the stability margin of the underlying attitude controller. This solution quantifies this constraint and generates hard constraints through offline safety boundary modeling.

[0025] 1. Parameter extraction and CFD flow field analysis: A three-dimensional geometric model and initial aerodynamic performance parameters of the UAV flight operation platform were obtained. These initial aerodynamic performance parameters specifically include: the maximum hovering thrust of the flight operation platform, the extreme speed envelope of a single motor, the basic lift-drag curve of the propeller blades, and the rotational inertia parameters of the airframe. These parameters are fundamental to accurately reproducing the effects of confined flow fields. While preserving the characteristics of the arms, blades, and main airframe, the three-dimensional geometric model was reasonably simplified and subjected to polyhedral mesh generation.

[0026] Considering the unique aerodynamic characteristics of multirotors during near-plane flight, a slip mesh computational method from computational fluid dynamics (CFD) was employed. Boundary conditions for the computational domain were set (e.g., the wall was set as a non-slip wall boundary, and the far field as a pressure outlet), and the mesh independence and time step independence were rigorously verified. This revealed the transient aerodynamic performance of the platform under different confined environments (such as single-wall and double-sided passageways). Figure 2and Figure 3 As shown, when a drone approaches the ground at low and high speeds, the flow field morphology formed by the interference between its downwash and the ground differs significantly; simultaneously, as Figure 4 As shown, when the UAV is in a tilted attitude, the confined space induces asymmetric vortex accumulation, resulting in strong aerodynamic moment (M) disturbances. The results of transient numerical analysis of the platform at different approach distances are smoothed, and the fluctuation range and peak parameters of the aerodynamic forces (X / Y / Z axial forces) and aerodynamic moments (roll / pitch / yaw moments) on the platform are extracted and classified. This data is then packaged as the external disturbance input for subsequent closed-loop analysis.

[0027] 2. Closed-loop system simulation and stability assessment: The extracted aerodynamic forces and torque fluctuations are treated as real-time external disturbances and applied to a pre-established closed-loop control simulation model of the aircraft system (this closed-loop control simulation model embeds a cascaded PID or active disturbance rejection controller (ADRC) used in actual aircraft). The time-domain and frequency-domain response performance indicators of the closed-loop system under disturbance conditions are comprehensively considered.

[0028] Based on this, the criteria set for "safe flight status" is defined. : Set as the maximum overshoot of the closed-loop system Less than or equal to the maximum allowable overshoot Adjusting time Less than or equal to the maximum allowable adjustment time And the minimum gain margin to ensure the system does not become unstable. Phase margin .

[0029] Only when a flight state falls within this set of criteria will its control commands not be saturated, thus effectively suppressing aerodynamic disturbances.

[0030] 3. 3D envelope surface fitting and hard constraint transformation: Under a defined flight condition (preset speed and attitude), simulation is used to make the UAV gradually approach the physical boundary of the environment in the X, Y, and Z directions of the aircraft coordinate system. When the recording system is affected by aerodynamic disturbances at the corresponding distance, its comprehensive performance evaluation index drops below [a certain threshold] for the first time. Determine the critical approximation distance when setting up the set. This is done by traversing the typical total thrust and velocity v range and attitude in a confined space. The dynamic range is used to comprehensively quantify the critical state of the flight platform as it approaches a constrained environment. Finally, a surface fitting algorithm is used to construct a dynamic range that varies with the platform's velocity v and attitude. A dynamically changing three-dimensional spatial envelope surface function. For example... Figure 5 As shown, the thickness of the aerodynamic safety boundary dynamically scales with flight speed, and the area inside this envelope is a safe and controllable region; the boundary line represents the limit of the system's disturbance rejection capability. Figure 6 As shown, under a specific tilt attitude, this safety boundary undergoes asymmetric distortion towards the side prone to airflow instability. This envelope surface function serves as the output of the UAV's dynamic aerodynamic safety boundary hard constraint model. This represents the UAV's dynamic aerodynamic safety flight boundary. .

[0031] To ensure the integration of the planning and control layers, the quantized three-dimensional spatial envelope is transformed into a mathematical analytical expression that can be embedded in the rolling temporal trajectory optimization (MPC) framework. When generating the reference trajectory, the final explicit hard constraints that integrate aerodynamic characteristics are formed: Within the future prediction time domain N, the planner is required to predict spatial locations. relative to the perceived location of environmental obstacles The predicted distance between them not only meets the static geometric collision avoidance distance And it must be greater than or equal to the dynamic aerodynamic safety boundary thickness calculated by substituting the current predicted motion velocity and predicted attitude into the three-dimensional envelope surface function.

[0032] This hard constraint ensures that the generated trajectory avoids obstacles in space while strictly limiting its flight state within the capability envelope of the underlying control system to stably suppress aerodynamic disturbances, effectively improving flight safety in complex and confined environments.

[0033] S2: Acquire multi-source point cloud observation data collected by airborne heterogeneous sensors, convert the multi-source point cloud observation data to the local body coordinate system for fusion and noise reduction, extract the geometric features of the confined environment, and generate and output the confined environment risk state characterization vector. The core of environmental perception is to characterize the current and future degree of local spatial confinement in real time, and generate a digital risk status.

[0034] 1. Multi-source point cloud fusion and denoising: Establishing a local body coordinate system in unknown and constrained environments. Using the current center of mass of the UAV as the origin, the longitudinal axis of the aircraft as the X-axis, the vertically upward axis as the Z-axis, and the Y-axis determined by the right-hand rule. Obtain the external parameter calibration matrices (including rotation matrices) of various heterogeneous sensors such as binocular vision cameras, infrared radar, and millimeter-wave radar. With translation vector The observation point cloud in the independent coordinate system of each sensor. Unified transformation to local body coordinate system The following is represented as Subsequently, the multi-source point cloud observation data are aligned and fused, and data simplification and environmental denoising are performed based on spatial point density characteristics to eliminate non-physical obstacles such as dust or water mist blown by wind, thereby obtaining the local effective obstacle point cloud set at the current time k. .

[0035] 2. Extracting geometric features and constructing risk characterization: For "channel and gap type risks" in confined environments, the geometric features are extracted from the fused point cloud dataset. In the process, key geometric features of the confined environment are extracted along the predicted flight path of the UAV: Distance to nearest obstacle ahead : The Euclidean distance from the center of mass of the drone to the nearest obstacle point in the projected area directly in front of it; The horizontal width that can be passed in front The minimum lateral span between the boundary points of obstacles on the left and right sides within the detection distance ahead of the predicted trajectory; Vertical height for passage ahead The minimum vertical span between obstacles (such as ceiling and ruin floor) in front of the predicted trajectory.

[0036] To characterize the dynamic risk level of drones approaching confined spaces, a confined environment risk state representation vector is constructed, with "accessible width and height" and their rate of change as the core. : in, These correspond to the approximation or change trends of the aforementioned geometric dimensions, respectively. Using discrete time... The real-time rate of change can be calculated online using the backward difference within the interval: This state vector It will be directly used as the state input space of the reinforcement learning (RL Agent), enabling it to perceive whether the environment is "narrowing" or "widening", thereby adaptively adjusting the safety weights of trajectory planning.

[0037] S3: Construct a rolling time-domain trajectory optimization objective function including planning target weight parameters, use the restricted environment risk state representation vector as the state space input reinforcement learning adaptive parameter tuning model, calculate and output the smoothly updated planning target weight parameters; Traditional trajectory optimization uses fixed constant weight parameters, making it difficult to balance flight efficiency in open areas with absolute safety in confined spaces. This invention introduces reinforcement learning to achieve online parameter tuning.

[0038] In the Rolling Time Domain Optimization (MPC) framework, a trajectory optimization objective function J is defined within the planning time domain N. To balance flight requirements under different environments, the objective function is composed of a weighted sum of multiple sub-terms, mainly including: a cost term representing the trade-off between flight efficiency and trajectory tracking error. The cost term representing the smoothness of control commands (such as jumps and accelerations) And penalties representing the safety margin for obstacle avoidance. And configure initial planning target weight parameters for each sub-item ( ). In traditional methods, these weights are fixed constants; however, in this invention, they are dynamically generated as time-varying parameters by an online parameter tuning mechanism. The objective function is expressed as: The generated constrained environment risk state representation vector Directly used as the state space feature at the current time k The input is fed into a pre-trained reinforcement learning adaptive parameter tuning model (RL Agent, such as using the PPO algorithm). The RL Agent outputs the action space based on the current environment's channel width, approximation velocity, and surface uncertainties. The model outputs the target action values ​​for each planning objective weight parameter. .

[0039] To ensure the stability and robustness of the planning process, a mechanism for smoothing the update of the RL Agent's output is introduced, including amplitude limiting and low-pass filtering. in, This is a limiting function, restricting the parameters within a reasonable physical range; This is a smoothing coefficient. When a drone enters a high-risk, narrow passage ( (The display width decreases sharply), and the RL strategy will adaptively increase the value. and and reduce This would suppress aggressive maneuvering; conversely, it would increase efficiency weighting.

[0040] S4: The dynamic aerodynamic safety boundary hard constraint model and the geometric features of the confined environment are used together as constraints for optimization. The rolling time-domain trajectory optimization objective function is updated using the smoothly updated planning objective weight parameters. The reference trajectory is solved and output. The closed-loop executability of the reference trajectory is verified. If the preset safety boundary margin is met, the reference trajectory is output to the underlying flight control system for execution. If it is not met, the safety action avoidance strategy is triggered and avoidance control commands are output.

[0041] This step is responsible for generating the final flight control commands and establishing the last line of defense for safety.

[0042] In the rolling time-domain optimization framework, the dynamic aerodynamic safety boundary hard constraint model and the constrained environment geometric features are used together as constraints for optimization. The rolling time-domain trajectory optimization objective function is updated using the smoothly updated planning objective weight parameters to solve and output the reference trajectory. Specifically, the optimization solver (using sequential quadratic programming (SQP) or other nonlinear solving algorithms) applies dynamic aerodynamic safety boundary hard constraints within the physical travel space defined by the constrained environment geometric features. Under the premise of satisfying the UAV dynamic differential flatness characteristics and the physical extrema of the actuator, the updated objective function J is minimized, and the first state point of the optimization sequence is extracted to form the reference trajectory.

[0043] Within each control cycle, perform closed-loop executability verification: extract the initial spatial state points on the generated reference trajectory, and use airborne sensor point cloud data to obtain the actual lateral distance to obstacles at the current moment. Based on the current predicted velocity and attitude, the corresponding dynamic aerodynamic safety boundary thickness determined in step S1 is extracted. The safety boundary margin at the current moment is calculated by subtracting the dynamic aerodynamic safety boundary thickness from the actual lateral distance to the obstacle. : Perform logical checks: determine whether the optimization solver has successfully found a feasible solution, and the safety boundary margin. If the value is strictly greater than zero, then the generated trajectory is mathematically feasible and provides sufficient redundancy for aerodynamic disturbances. The verification is successful, and the reference trajectory is output to the underlying flight control system for execution.

[0044] In the event of extreme circumstances (such as sudden strong gusts of wind, complete blockage of the path ahead leading to an unsolvable objective function, or a calculated safety boundary margin less than or equal to zero, indicating impending aerodynamic instability and crash), the normal execution process is immediately interrupted, triggering a safety avoidance strategy. The planner takes over flight control commands and forcibly outputs emergency control actions to the underlying actuators: if the speed is low, a hovering command is output; if approaching an obstacle at high speed, an emergency braking maximum reverse thrust command is output; if encountering a dead end, a reverse control command is output along the previously verified safe historical trajectory. The avoidance strategy is lifted and autonomous planning and solving resume only after the system reassesses and finds that the newly extracted restricted environment risk state representation vector has recovered to within the preset safety threshold range.

[0045] In summary, the UAV adaptive rolling trajectory planning method integrating aerodynamic safety boundary constraints provided by this invention possesses complete aerodynamic mechanism support and highly intelligent environmental adaptive parameter tuning capabilities, and is equipped with rigorous safety boundary verification and action avoidance mechanisms. It can significantly enhance the robustness, autonomous obstacle avoidance capability, and mission execution reliability of UAVs flying in extremely narrow and confined spaces such as underground utility tunnels and rubble gaps.

[0046] Example 2 Please see Figure 7 The diagram shown is a structural schematic of an adaptive rolling trajectory planning system for unmanned aerial vehicles (UAVs) according to a second embodiment of this application. The system includes the following key modules: The dynamic aerodynamic safety boundary hard constraint modeling module 100 is used to obtain the three-dimensional geometric model and initial aerodynamic performance indicators of the UAV flight operation platform, construct the system closed-loop control simulation model, obtain the flow field disturbance through computational fluid dynamics and input it into the system closed-loop control simulation model, and generate and output the UAV dynamic aerodynamic safety boundary hard constraint model. The confined environment perception and risk characterization module 200 is used to acquire multi-source point cloud observation data collected by airborne heterogeneous sensors, convert the multi-source point cloud observation data to the local body coordinate system for fusion and denoising, extract the geometric features of the confined environment, and generate and output the confined environment risk state characterization vector. The reinforcement learning adaptive planning parameter tuning module 300 is used to construct a rolling time-domain trajectory optimization objective function including planning target weight parameters, and to input the restricted environment risk state representation vector as the state space into the reinforcement learning adaptive parameter tuning model, and to calculate and output the smoothly updated planning target weight parameters. The rolling trajectory optimization solution module 400 is used to take the dynamic aerodynamic safety boundary hard constraint model and the geometric features of the confined environment as the constraint conditions for optimization solution, and use the smoothly updated planning target weight parameters to update the rolling time domain trajectory optimization objective function, solve and output the reference trajectory; The safety verification execution module 500 is used to perform closed-loop executability verification on the reference trajectory. If the preset safety boundary margin is met, the reference trajectory is output to the underlying flight control system for execution. If it is not met, a safety action avoidance strategy is triggered to output an avoidance control command.

[0047] The UAV adaptive rolling trajectory planning system in this application embodiment can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network-attached storage (NAS), personal computers (PCs), etc. This application embodiment does not impose specific limitations.

[0048] The UAV adaptive rolling trajectory planning system in this application embodiment can be a device with an operating system. The operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit the specific operating system.

[0049] The UAV adaptive rolling trajectory planning system provided in this application embodiment can achieve... Figure 1 The various processes of an adaptive rolling trajectory planning method for unmanned aerial vehicles (UAVs) implemented in the method embodiments are not described in detail here to avoid repetition.

[0050] Optionally, embodiments of this application also provide an electronic device, including a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the various processes of the above-described embodiment of the UAV adaptive rolling trajectory planning method and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0051] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described embodiment of the UAV adaptive rolling trajectory planning method and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0052] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0053] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0054] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, 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 (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0055] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for adaptive rolling trajectory planning of unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: The three-dimensional geometric model and initial aerodynamic performance indicators of the UAV flight operation platform are obtained, a system closed-loop control simulation model is constructed, flow field disturbances are obtained through computational fluid dynamics and input into the system closed-loop control simulation model, and a dynamic aerodynamic safety boundary hard constraint model of the UAV is generated and output. Acquire multi-source point cloud observation data collected by airborne heterogeneous sensors, transform the multi-source point cloud observation data to the local body coordinate system for fusion and denoising, extract the geometric features of the confined environment, and generate and output the confined environment risk state characterization vector. A rolling time-domain trajectory optimization objective function including planning target weight parameters is constructed. The constrained environment risk state representation vector is used as the state space input reinforcement learning adaptive parameter tuning model to calculate and output the smoothly updated planning target weight parameters. The dynamic aerodynamic safety boundary hard constraint model and the geometric features of the confined environment are used together as constraints for optimization, and the rolling time-domain trajectory optimization objective function is updated using the smoothly updated planning objective weight parameters to solve and output the reference trajectory. The reference trajectory is subjected to closed-loop executability verification. If the preset safety boundary margin is met, the reference trajectory is output to the underlying flight control system for execution. If it is not met, a safety action avoidance strategy is triggered and an avoidance control command is output.

2. The method according to claim 1, characterized in that, Generate and output a dynamic aerodynamic safety boundary hard constraint model for the UAV, specifically including: Based on the initial aerodynamic performance indicators, the three-dimensional geometric model of the UAV flight operation platform is meshed. The transient values ​​under different constrained environments and at different approximation distances are calculated using the sliding mesh calculation method. The fluctuation range and peak parameters of the platform's aerodynamic forces and aerodynamic torques are extracted as real-time external disturbances. The external real-time disturbance is input into the closed-loop control simulation model of the system to obtain the time-domain response performance index and frequency-domain response performance index under the disturbance condition. The time-domain response performance index includes the maximum overshoot and the settling time, and the frequency-domain response performance index includes the amplitude margin and the phase margin. A safety performance judgment set is constructed, wherein the safety performance judgment set is set as follows: the maximum overshoot of the closed-loop system is less than or equal to the maximum allowable overshoot, the settling time is less than or equal to the longest allowable settling time, the amplitude margin is greater than or equal to the minimum amplitude margin, and the phase margin is greater than or equal to the minimum phase margin. Under the set flight conditions, the aircraft gradually approaches the physical boundary of the environment in each direction of the body coordinate system, and records the critical approach distance when the comprehensive performance evaluation index first falls out of the safety performance judgment set. By traversing the set total thrust and the speed and attitude ranges under confined space, the critical state of the UAV flight operation platform when approaching the constrained environment is quantified, and a three-dimensional spatial envelope surface function that dynamically changes with the platform's motion speed and attitude is fitted and constructed, and the output is used as the hard constraint model of the UAV's dynamic aerodynamic safety boundary.

3. The method according to claim 2, characterized in that, The dynamic aerodynamic safety boundary hard constraint model transforms its quantized three-dimensional spatial envelope into a mathematical analytical expression embedded in the rolling time-domain trajectory optimization, forming explicit hard constraint conditions. These explicit hard constraint conditions are: Within the rolling prediction time domain, the predicted distance between the UAV's future predicted spatial position and the perceived position of environmental obstacles not only satisfies the static geometric collision avoidance distance, but the predicted distance is also greater than or equal to the dynamic aerodynamic safety boundary thickness calculated by substituting the current predicted motion speed and predicted attitude into the three-dimensional spatial envelope surface function.

4. The method according to claim 1, characterized in that, Acquire multi-source point cloud observation data from airborne heterogeneous sensors, transform the multi-source point cloud observation data to a local body coordinate system for fusion and denoising, and extract geometric features of the confined environment, specifically including: With the current center of mass of the UAV as the origin, the longitudinal axis of the UAV body as the longitudinal axis, the vertical upward as the vertical axis, and the horizontal axis determined by the right-hand rule, a local coordinate system is established. Obtain the external parameter calibration matrix of each of the airborne heterogeneous sensors. The external parameter calibration matrix includes a rotation matrix and a translation vector. The multi-source point cloud observation data of the airborne heterogeneous sensors in their respective coordinate systems are uniformly transformed to the local body coordinate system through the external parameter calibration matrix. The transformed multi-source point cloud observation data are aligned and fused, and data simplification and environmental denoising are performed based on spatial point density characteristics to extract the local effective obstacle point cloud set at the current moment. In the set of local effective obstacle point clouds, the geometric features of the restricted environment are extracted along the direction of the UAV's predicted flight trajectory. The geometric features of the restricted environment include, in sequence, the distance to the nearest obstacle in front, the passable lateral width in front, and the passable vertical height in front.

5. The method according to claim 4, characterized in that, Generate and output a constrained environment risk state representation vector, specifically including: The sampling period within the discrete time is obtained. The difference between the current time and the previous time is calculated using backward difference and divided by the sampling period to obtain the real-time rate of change of the forward passable lateral width. The difference between the current and previous passable vertical heights is calculated using backward difference and divided by the sampling period to obtain the real-time rate of change of the passable vertical heights. The current passable lateral width, the real-time rate of change of the passable lateral width, the passable vertical height, and the real-time rate of change of the passable vertical height are combined in a set order to construct, generate, and output the restricted environment risk state representation vector.

6. The method according to claim 1, characterized in that, Calculate and output the smoothly updated planning objective weight parameters, specifically including: The objective function for the rolling time-domain trajectory optimization is set to include a cost term representing flight efficiency and trajectory tracking error, a cost term representing the smoothness of control commands, and a penalty term representing obstacle avoidance safety margin, and corresponding initial planning objective weight parameters are configured for each of the cost terms and penalty terms. The generated risk state representation vector of the restricted environment is used as the state space feature at the current moment and input into the reinforcement learning adaptive parameter tuning model. The reinforcement learning adaptive parameter tuning model performs network forward propagation calculation based on the channel width and approximation speed change characteristics of the current restricted environment, and outputs the target action value of each planning target weight parameter. By using preset upper and lower limit functions, each target action value is truncated and restricted within a preset physical operation range to obtain the limited weight parameters. Based on the set smoothing coefficient, the planned target weight parameters of the previous control time and the weight parameters after the current time limit are weighted and summed to calculate and output the final smoothed updated planned target weight parameters.

7. The method according to claim 1, characterized in that, The closed-loop executability verification of the reference trajectory specifically includes: Extract the initial spatial state point on the generated reference trajectory and use airborne sensor data to obtain the actual lateral distance to the obstacle at the current moment; Based on the current flight status, the corresponding dynamic aerodynamic safety boundary thickness is extracted. The actual obstacle lateral distance is subtracted from the dynamic aerodynamic safety boundary thickness to calculate the safety boundary margin at the current moment. Determine whether the optimization solver has successfully found a feasible solution and whether the safety boundary margin is greater than zero; If the solver successfully finds a feasible solution and the safety boundary margin is greater than zero, the verification is successful, and the reference trajectory is output to the underlying flight control system for execution. If the objective function for optimizing the rolling time-domain trajectory has no solution, or if the safety boundary margin is less than or equal to zero, a safety action retreat strategy is triggered. The planner takes over the flight control commands and outputs retreat control commands to the actuators, including hovering, emergency braking, or retreating along the original safety trajectory, until the latest extracted restricted environment risk state representation vector is restored to within the preset safety threshold range.

8. An adaptive rolling trajectory planning system for unmanned aerial vehicles (UAVs), characterized in that, include: The dynamic aerodynamic safety boundary hard constraint modeling module is used to obtain the three-dimensional geometric model and initial aerodynamic performance indicators of the UAV flight operation platform, construct the system closed-loop control simulation model, obtain the flow field disturbance through computational fluid dynamics and input it into the system closed-loop control simulation model, and generate and output the UAV dynamic aerodynamic safety boundary hard constraint model. The confined environment perception and risk characterization module is used to acquire multi-source point cloud observation data collected by airborne heterogeneous sensors, convert the multi-source point cloud observation data to the local body coordinate system for fusion and noise reduction, extract the geometric features of the confined environment, and generate and output the confined environment risk state characterization vector. The reinforcement learning adaptive planning parameter tuning module is used to construct a rolling time-domain trajectory optimization objective function including planning target weight parameters, and to input the restricted environment risk state representation vector as the state space into the reinforcement learning adaptive parameter tuning model, and to calculate and output the smoothly updated planning target weight parameters. The rolling trajectory optimization solution and safety verification execution module is used to take the dynamic aerodynamic safety boundary hard constraint model and the geometric features of the confined environment as the constraint conditions for optimization solution, and use the smoothly updated planning target weight parameters to update the rolling time domain trajectory optimization objective function, solve and output the reference trajectory; The reference trajectory is subjected to closed-loop executability verification. If the preset safety boundary margin is met, the reference trajectory is output to the underlying flight control system for execution. If it is not met, a safety action avoidance strategy is triggered and an avoidance control command is output.

9. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein when the program or instructions are executed by the processor, they implement the steps of the UAV adaptive rolling trajectory planning method as described in any one of claims 1-7.

10. A readable storage medium, characterized in that, The program or instructions are stored on the readable storage medium, and when the program or instructions are executed by the processor, they implement the steps of the UAV adaptive rolling trajectory planning method as described in any one of claims 1-7.

Images