An adaptive smooth obstacle avoidance braking method and device for medium and large unmanned aerial vehicles

Abstract

The application discloses a kind of adaptive smooth obstacle avoidance braking method and equipment for medium and large unmanned aerial vehicle.The method comprises: after time synchronization of obstacle observation data and unmanned aerial vehicle operating state data, the relative motion state of obstacle is fused and estimated based on the state of unmanned aerial vehicle itself, the collision time estimate value and its standard deviation are calculated;According to the total mass of unmanned aerial vehicle, the maximum deceleration is determined and two dynamic thresholds of emergency braking and smooth braking are calculated;The safe modified collision time is obtained by using the collision time estimate value and the standard deviation, and whether to brake and brake mode is decided by comparing with two thresholds;When smooth braking, stiffness coefficient is determined according to mass and continuous growth deceleration command is generated according to power function, when emergency braking, optimal deceleration command is solved by constrained optimization taking smooth braking reference trajectory as target;The command is converted into control output and feedback operating state data to realize closed loop.The application also proposes corresponding equipment, and the application realizes full load adaptive smooth braking of unmanned aerial vehicle.

Description

Technical Field

[0001] This invention belongs to the field of smooth obstacle avoidance braking control technology for unmanned aerial vehicles (UAVs), and specifically relates to an adaptive smooth obstacle avoidance braking method and device for medium and large UAVs. Background Technology

[0002] With the rapid development of drone technology, medium and large-sized drones are playing an important role in industrial applications such as logistics delivery, power line inspection, and emergency rescue. Most existing drone obstacle avoidance systems are designed for consumer-grade small drones and generally employ a simple strategy of triggering braking based on a fixed distance threshold; that is, an emergency stop is triggered when the distance between the drone and an obstacle is less than a preset threshold. This strategy is acceptable for small drones, but medium and large-sized multi-rotor drones weighing 12kg and above have significant characteristics such as large payload, large inertia, and slow response of the power system, posing more stringent technical challenges to their obstacle avoidance and braking systems.

[0003] From the perspective of technological evolution, UAV obstacle avoidance technology has undergone three generations of development: the first generation is based on passive obstacle avoidance at close range using a single ultrasonic or infrared sensor; the second generation is based on active perception combined with fixed threshold braking using vision or lidar; and the third generation is the currently developing multi-source fusion perception and intelligent decision-making braking. The existing third-generation technology mainly focuses on the optimization of environmental perception and path planning. For the braking control link, especially for the smoothness and adaptability issues of medium and large UAVs with large inertia platforms, there is still a lack of systematic solutions. The existing technology has the following shortcomings: (1) The fixed threshold method has poor adaptability. Patent CN111722640B discloses that obstacle avoidance actions are triggered by setting a fixed absolute distance threshold. This method cannot adapt to changes in different loads and flight speeds. That is, insufficient braking distance at high speeds or heavy loads can easily lead to collisions, and premature braking at low speeds or light loads affects operational efficiency. Its fundamental defect is that it treats the braking distance requirement as a constant and ignores the physical law that the braking distance is proportional to the square of the speed and positively correlated with the mass. (2) The general flight control algorithm is not optimized for large inertia. When the standard control algorithms of open-source flight control platforms such as PX4 are used to handle large inertia UAVs of 12kg class and above, the position loop step command directly generates step acceleration output, resulting in excessive impact during braking; there is a lack of adaptive adjustment mechanism based on real-time quality, and different loads require manual readjustment of parameters; sudden acceleration changes can easily lead to attitude divergence or even instability, and the high-frequency impact generated at the same time accelerates the fatigue of the fuselage structure and damages precision loads. (3) Limited ability to deal with dynamic obstacles. Existing systems mostly adopt a switch-type emergency stop strategy, and the discontinuous control quantity leads to violent fluctuations in pitch angle; at the same time, the judgment is based on absolute distance rather than collision time, and the obstacle speed information is not integrated, resulting in a lag in response to high-speed obstacles moving in opposite directions. (4) Insufficient accuracy and fault tolerance of multi-source sensor fusion. The sampling frequencies of lidar, vision camera and millimeter-wave radar are different and the observation dimensions are not uniform. The existing weighted fusion scheme has low time synchronization accuracy and lacks a unified fault detection and isolation mechanism. When a single sensor fails, the overall system is severely degraded.

[0004] In summary, there is an urgent need for a drone obstacle avoidance braking scheme that can adjust the braking timing in real time according to flight speed and drone mass, adapt to different load conditions, achieve smooth deceleration transition, and has high fault-tolerant fusion capability of multi-source sensors. Summary of the Invention

[0005] To address the aforementioned technical challenges, this invention proposes an adaptive smooth obstacle avoidance braking method and device for medium-to-large-sized unmanned aerial vehicles (UAVs). Through TTC dynamic hierarchical decision-making, mass adaptive stiffness matching, power function smooth braking, and high-fault-tolerant fusion of multi-source sensors, adaptive smooth braking of the UAV is achieved across the entire payload range and at different flight speeds. While ensuring a safe braking distance, it effectively suppresses structural impact and eliminates attitude overshoot.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, this invention proposes an adaptive smooth obstacle avoidance and braking method for medium and large-sized unmanned aerial vehicles (UAVs), comprising the following steps: After synchronizing the acquired obstacle observation data and UAV operating status data, the synchronized UAV operating status data is used as a known reference quantity. The relative motion state of the obstacle is estimated by fusing the data through a state estimation filtering algorithm. The collision time estimate and the estimated standard deviation of the collision time estimate are calculated based on the relative motion state. Based on the current total mass of the UAV, the maximum available deceleration is determined. A first threshold for triggering emergency braking and a second threshold for triggering smooth braking are calculated based on the current flight speed, the maximum available deceleration, and the total delay from detection to braking. The first threshold is less than the second threshold. Based on the estimated collision time and the estimated standard deviation, a safety correction collision time is determined. The safety correction collision time is compared with a first threshold and a second threshold: if the safety correction collision time is greater than the second threshold, braking is not triggered and the system returns to reacquire obstacle observation data and UAV operational status data; if the safety correction collision time is between the first threshold and the second threshold, a smooth braking mode is entered; if the safety correction collision time is not greater than the first threshold, an emergency braking mode is entered. In smooth braking mode, the braking stiffness coefficient is determined based on the current total mass of the UAV. A continuously increasing deceleration command is generated according to the power function law, using the braking stiffness coefficient as a parameter and the difference between the second threshold and the current collision time estimate as the time margin. The deceleration command value when it drops to the first threshold is used as the reference trajectory. In emergency braking mode, the optimal deceleration command is solved by constraint optimization with the reference trajectory as the target. The optimal deceleration command is converted into pitch angle control and total thrust control, and output to the flight control system to perform braking. At the same time, the current operating status data of the UAV is acquired and returned to participate in the state estimation and closed-loop correction of the next cycle.

[0007] Furthermore, the detailed process of synchronizing the acquired obstacle observation data and UAV operational status data in time is as follows: Obstacle observation data acquisition: Obstacle distance and azimuth angle are collected using airborne lidar; depth maps of obstacles are collected using airborne visual cameras; radial velocity data of obstacles are collected using airborne millimeter-wave radar. Acquire UAV operational status data: The flight control system outputs the UAV's current position, speed, and attitude in real time; the inertial measurement unit collects the three-axis acceleration and three-axis angular velocity of the UAV in real time. The obstacle observation data and the UAV operation status data are each attached with a timestamp in the form of data frames. Based on the timestamps of each data, the data with different sampling frequencies are interpolated and aligned to a unified fusion time to complete the time synchronization of the data.

[0008] Furthermore, using the synchronized UAV operational status data as a known baseline, the relative motion state of the obstacle is estimated through a state estimation filtering algorithm. The collision time estimate and its standard deviation are then calculated based on the relative motion state. Specifically: Using the position, velocity, and attitude in the synchronized UAV operating state data as known reference inputs, and the relative position and velocity of the obstacle relative to the UAV as state variables to be estimated, the state equation of the state estimation filtering algorithm is established. The obstacle distance and azimuth angle observed by lidar, the obstacle depth map observed by visual camera, and the radial velocity observed by millimeter-wave radar are respectively established as corresponding observation equations; The observation data is sequentially updated according to the arrival order, and the relative position and relative velocity of the obstacle relative to the UAV are obtained by fusion estimation; The collision time estimate is calculated based on the ratio of the relative position to the relative velocity, and the estimated standard deviation of the collision time estimate is calculated based on the state covariance matrix output by the state estimation filtering algorithm.

[0009] Furthermore, based on the current flight speed, the maximum available deceleration, and the total delay from detection to braking, a first threshold for triggering emergency braking and a second threshold for triggering smooth braking are calculated, specifically as follows: The first threshold is the sum of a first set ratio of the pure deceleration time and the total delay; the pure deceleration time is the ratio of the current flight speed to the maximum available deceleration. The second threshold is the sum of a second set ratio of the pure deceleration time and the total delay; and the second set ratio is greater than the first set ratio.

[0010] Furthermore, the safety-corrected collision time is determined by subtracting a first set multiple of the estimated standard deviation from the estimated collision time, and the resulting difference is taken as the safety-corrected collision time.

[0011] Furthermore, the process of generating the deceleration command is as follows: ; in, Indicates a deceleration command; Indicates the braking stiffness coefficient; Indicates the second threshold; This represents the estimated collision time. This represents the preset normalized time constant; This represents the smoothing factor.

[0012] Furthermore, in the emergency braking mode, the optimal deceleration command is solved through constraint optimization with the reference trajectory as the target, specifically as follows: Using the reference trajectory as the deceleration tracking target, a multi-objective cost function is constructed that includes deceleration tracking error, attitude angle deviation, and thrust change rate. The constraints are limited by the speed of the UAV motor, the pitch angle, and the rate of change of thrust. Within each control cycle, with the goal of minimizing the cost function, the optimal control sequence in the preset time domain is solved within the feasible domain defined by the constraints, and the first control variable in the optimal control sequence is output as the optimal deceleration command for the current cycle.

[0013] Furthermore, methods for obtaining the current total mass of the drone include: During the takeoff and hovering phase, the initial mass is calculated by back-calculating the throttle and thrust calibration curves. During the level flight phase, the mass estimate is updated online based on the longitudinal thrust balance equation. During braking, the sliding average of the most recent valid mass estimate before braking is triggered is used as the current total mass of the UAV, and the mass change is compensated by integrating the battery current.

[0014] Furthermore, the method also includes: when the relative velocity of the obstacle is detected to be less than or equal to zero and the relative distance to the obstacle continues to increase in multiple consecutive frames, it is determined that the collision risk has been eliminated, and the deceleration command exits the braking process according to an exponential decay law; the exponential decay law is specifically as follows: ; in, This indicates the deceleration command at the current moment; The deceleration command indicating exit from the start time. Indicates the current moment; Indicates the start time of exit; This represents the decay time constant.

[0015] Secondly, the present invention also proposes an adaptive smooth obstacle avoidance braking device for medium and large-sized unmanned aerial vehicles (UAVs), comprising at least one processor and a memory, wherein the memory stores a computer program, and the computer program, when executed by the at least one processor, implements the aforementioned adaptive smooth obstacle avoidance braking method for medium and large-sized UAVs.

[0016] The effects described in the invention are merely those of the embodiments, and not all the effects of the invention. One of the above technical solutions has the following advantages or beneficial effects: (a) Synergistic improvement of braking safety and operational efficiency This invention abandons the traditional coarse-grained decision-making method based on fixed distance thresholds and adopts a dynamic hierarchical decision-making mechanism based on collision time. By fusing velocity observation information from millimeter-wave radar, it achieves accurate perception and timely response to dynamic obstacles. When facing high-speed obstacles moving in opposite directions, this invention can trigger braking at an earlier time, reserving sufficient safety margin for the UAV; while when the obstacle is relatively stationary or far away, the braking triggering timing is correspondingly delayed, minimizing unnecessary mission interruptions. Compared with traditional fixed threshold schemes, this invention significantly improves the safety margin in dynamic obstacle scenarios. Furthermore, by incorporating the estimated standard deviation of the collision time into the calculation of the safety-corrected collision time, the system automatically adopts a more conservative braking strategy when the perception confidence decreases, further ensuring flight safety.

[0017] (ii) Adaptive braking capability across the entire load range The mass-adaptive stiffness model introduced in this invention automatically adjusts the braking system's response characteristics according to the total mass of the UAV. Under heavy load conditions, the system automatically increases braking stiffness to ensure sufficient braking deceleration and braking distance margin; under light load conditions, the system automatically reduces braking stiffness to avoid impacts and attitude disturbances caused by over-braking. This mechanism enables the invention to adapt to the full payload range of medium and large UAVs without requiring manual parameter adjustments for different payloads, simplifying the usage process and avoiding safety risks caused by improper parameter tuning. A three-level strategy is employed for mass acquisition, effectively avoiding the coupling effect of acceleration disturbances during braking on mass estimation, ensuring the accuracy and stability of mass parameters under different operating conditions.

[0018] (III) Effective protection of fuselage structure and precision loads This invention employs a power function model to generate smooth deceleration commands, eliminating step responses at the source of control input. In the initial braking phase, the deceleration increases extremely gradually, avoiding damage to the fuselage structure from instantaneous impacts and disturbances to precision loads such as the gimbal and optoelectronic pods. As the collision time decreases, the deceleration increases at an incremental rate, ensuring sufficient braking force near the point of impact. Throughout the braking process, the fuselage pitch angle increases smoothly in an approximately S-shaped curve, without overshoot or oscillation, maintaining a stable and controllable attitude. Simultaneously, in emergency braking mode, the optimal deceleration command is solved through constraint optimization, achieving the fastest deceleration while satisfying multiple constraints such as motor speed, pitch angle, and thrust rate of change. This ensures safety and avoids the risks of attitude loss of control and power saturation.

[0019] (iv) High precision and high fault tolerance of multi-source sensing This invention employs a state estimation filtering algorithm to unify and fuse observation data from three heterogeneous sensors: lidar, visual cameras, and millimeter-wave radar. This fully leverages the complementary advantages of each sensor in distance, angle, and velocity observations. The observation noise covariance of each sensor adaptively adjusts with the measurement distance; at long distances, the fusion weight automatically decreases, effectively suppressing the contamination of the fusion result by long-distance observation noise. Simultaneously, sensor faults are detected in real-time using a chi-square test of the innovation sequence. When a sensor fails, its observation weight continuously and smoothly decreases to zero, achieving shock-free degradation of the fusion system. Single-point failures do not lead to the overall loss of system functionality, ensuring the system's basic braking capability even under partial sensor failure conditions.

[0020] (v) Smooth braking disengagement and efficient task recovery When the obstacle deviates or moves away, eliminating the collision risk, this invention smoothly releases the braking force using an exponential decay law. The deceleration command transitions continuously from its current value to zero, and the pitch angle synchronously and smoothly returns to a horizontal state. This exit method avoids fuselage rebound and attitude shock caused by sudden release of braking force, enabling the UAV to quickly and smoothly return to normal flight. Combined with the hysteresis mechanism in the hierarchical decision-making process to effectively suppress mode switching jitter, this invention maintains high smoothness and stability throughout the entire braking triggering, execution, and exit process, and also exhibits good robustness to external disturbances such as strong winds. Attached Figure Description

[0021] Figure 1 This is a flowchart of an adaptive smooth obstacle avoidance and braking method for medium and large unmanned aerial vehicles (UAVs) proposed in Embodiment 1 of the present invention. Figure 2 This is a schematic diagram of an adaptive smooth obstacle avoidance braking device for medium and large-sized UAVs proposed in Embodiment 2 of the present invention. Detailed Implementation

[0022] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure of the invention, components and arrangements of specific examples are described below. Furthermore, reference numerals and / or letters may be repeated in different examples. This repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. Descriptions of well-known components, processing techniques, and processes are omitted in this invention to avoid unnecessarily limiting the invention.

[0023] Example 1 Embodiment 1 of this invention proposes an adaptive smooth obstacle avoidance braking method for medium and large UAVs. This method addresses the problems in existing UAV obstacle avoidance braking systems, such as mismatch between braking timing and braking demand, poor adaptability under different loads, large impacts on the fuselage structure and precision loads, and decreased system reliability when some sensors fail. These problems arise because existing UAV obstacle avoidance braking systems suffer from issues such as fixed threshold decision-making, lack of mass adaptation, sudden deceleration steps, and insufficient accuracy and fault tolerance of multi-source sensor fusion.

[0024] This invention replaces the traditional fixed distance threshold with Time-to-Collision (TTC) as a unified decision quantity; it introduces real-time mass parameters of the UAV to construct an adaptive braking stiffness model; it uses a power function-type deceleration generation curve to achieve smooth braking transition; and it combines error state extended Kalman filtering with multi-source data fusion to improve perception accuracy and robustness.

[0025] Figure 1 This is a flowchart of an adaptive smooth obstacle avoidance and braking method for medium and large unmanned aerial vehicles (UAVs) proposed in Embodiment 1 of the present invention. In step S1, obstacle observation data and UAV operating status data are acquired.

[0026] Specifically, the point cloud distance and azimuth data of obstacles are collected by airborne lidar; the lidar is selected as a 16-line or 32-line mechanical rotating or solid-state lidar with a range of 0.5m to 100m, an accuracy of ±2cm, a horizontal angular resolution of 0.1° to 0.4°, and a sampling frequency of 10Hz to 20Hz.

[0027] Depth image data of obstacles are acquired through an onboard vision camera; the vision camera is a binocular global shutter camera with a baseline of 120mm to 200mm, an effective range of 0.5m to 30m, and a sampling frequency of 15Hz to 30Hz.

[0028] The distance and radial velocity data of obstacles are collected by an airborne millimeter-wave radar. The selected millimeter-wave radar is a 77GHz FMCW radar with a range of 1m to 80m, a distance accuracy of ±0.1m, a radial velocity accuracy of ±0.1m / s, and a sampling frequency of 20Hz to 50Hz. The above data together constitute the obstacle observation data.

[0029] Meanwhile, the flight control system outputs the UAV's current position, speed, and attitude in real time; the position, speed, and attitude are output by the flight control master extended Kalman filter and serve as the main components of the UAV's operational status data.

[0030] The inertial measurement unit (IMU) collects the three-axis acceleration and three-axis angular velocity of the drone in real time. A six-axis or nine-axis high-precision IMU is selected, with a sampling frequency of 100Hz to 500Hz. This data collectively constitutes the drone's operational status data.

[0031] Each sensor outputs the acquired data in the form of data frames with precise timestamps for subsequent time synchronization processing.

[0032] In step S2, after synchronizing the acquired obstacle observation data and UAV operating status data, the synchronized UAV operating status data is used as a known reference quantity. The relative motion state of the obstacle is estimated by fusing the data through a state estimation filtering algorithm. The collision time estimate and the estimated standard deviation of the collision time estimate are calculated based on the relative motion state. The process of synchronizing the acquired obstacle observation data and UAV operation status data is as follows: the obstacle observation data and the UAV operation status data are each attached with a timestamp in the form of data frames. Based on the timestamps of each data, the data with different sampling frequencies are interpolated and aligned to a unified fusion time to complete the time synchronization of the data.

[0033] Specifically, the obstacle observation data and UAV operational status data acquired in step S1 are synchronized in time. A two-level synchronization mechanism is adopted, consisting of hardware second pulse marking and timestamp interpolation alignment. All sensors are triggered for acquisition or marking by the same hardware second pulse signal, ensuring a unified hardware-level time reference. When data from each sensor arrives at the fusion module, it is based on a unified fusion time within the fusion cycle. The data is then interpolated to the unified fusion time through integration and extrapolation using the inertial measurement unit based on the precise timestamp, thus completing soft synchronization. The main fusion frequency is 50Hz, aligned with the millimeter-wave radar sampling frequency. Data from the lidar and vision systems is updated asynchronously upon arrival.

[0034] Then, using the position, velocity, and attitude in the synchronized UAV operating state data as known reference inputs, and the relative position and relative velocity of the obstacle relative to the UAV as state variables to be estimated, the state equation of the state estimation filtering algorithm is established. The obstacle distance and azimuth angle observed by lidar, the obstacle depth map observed by visual camera, and the radial velocity observed by millimeter-wave radar are respectively established as corresponding observation equations; The observation data is sequentially updated according to the arrival order, and the relative position and relative velocity of the obstacle relative to the UAV are obtained by fusion estimation; The collision time estimate is calculated based on the ratio of the relative position to the relative velocity, and the estimated standard deviation of the collision time estimate is calculated based on the state covariance matrix output by the state estimation filtering algorithm.

[0035] Collision time estimate The calculation formula is:

[0036] in, Indicates relative position; This indicates relative velocity.

[0037] Should Used for trend judgment of risk level. Based on the state covariance matrix output by the state estimation filtering algorithm, the estimated variances of the relative position and the relative velocity are extracted, and the estimated standard deviation of the collision time estimate is calculated according to the error propagation formula. .

[0038] Preferably, the state estimation filtering algorithm in this invention is an error state extended Kalman filter. The observation noise covariance of each sensor's observation data adaptively increases with increasing measurement distance; noise increases and fusion weight decreases at long distances, while noise decreases and fusion weight increases at short distances. Sensor faults are detected in real time using the chi-square test of the innovation sequence. When a sensor failure is detected, the observation weight of that sensor continuously and smoothly decreases to zero, achieving shockless degradation of the fusion system. The fusion output includes relative position. Relative velocity Collision time Estimated standard deviation Obstacle azimuth and health status flags of each sensor .

[0039] In step S3, the maximum available deceleration is determined based on the current total mass of the UAV. A first threshold for triggering emergency braking and a second threshold for triggering smooth braking are calculated based on the current flight speed, the maximum available deceleration, and the total delay from detection to braking. The first threshold is less than the second threshold. First, the current total mass of the UAV is obtained. This total mass is acquired through a three-level strategy: takeoff calibration, online estimation during the constant speed phase, and maintenance during braking. During takeoff and hovering, the initial mass is inversely estimated using the throttle and thrust calibration curves, achieving an accuracy of ±2%. During constant speed level flight, when the longitudinal acceleration is less than 0.2 m / s², online estimation and updates are performed based on the longitudinal thrust balance equation, and mass changes caused by battery consumption are slowly compensated for through current integration. During braking, mass is treated as a slow variable, and the sliding average of the most recent effective mass estimate before braking is used as the current total mass of the UAV, avoiding coupled oscillations between braking acceleration and mass estimation.

[0040] The maximum available deceleration is determined based on the total mass. The maximum usable deceleration is derived from the relationship between the maximum thrust, maximum allowable pitch angle, and total mass of the UAV power system, with typical values ​​for medium and large multi-rotor aircraft. to .

[0041] Current flight speed With maximum available deceleration The ratio is denoted as the pure deceleration time. The pure deceleration time represents the theoretically shortest time required to decelerate from the current speed to zero at maximum deceleration. The first threshold. The first set ratio of the pure deceleration time to the total delay from detection to braking. The sum of the second threshold The second set ratio of the pure deceleration time to the total delay The sum of the two, and the second set ratio is greater than the first set ratio, thereby ensuring .

[0042] Wherein, the total delay from detection to braking The total system delay includes the sum of sensor sampling period, data processing time, and motor response delay, and is obtained based on the specific hardware platform calibration.

[0043] Preferably, the first set ratio is 0.5, and the second set ratio is 1, that is: ; .

[0044] In step S4, a safe corrected collision time is determined based on the estimated collision time and the estimated standard deviation. The safe corrected collision time is compared with a first threshold and a second threshold: if the safe corrected collision time is greater than the second threshold, braking is not triggered and the system returns to reacquire obstacle observation data and UAV operational status data; if the safe corrected collision time is between the first threshold and the second threshold, the system enters a smooth braking mode and executes step S5; if the safe corrected collision time is not greater than the first threshold, the system enters an emergency braking mode and executes step S6. The method for determining the safety correction collision time is as follows: the estimated collision time is... Subtract the estimated standard deviation The first set multiple, the resulting difference is used as the safety correction collision time. ,Right now ; The corrected collision time is used to replace the original collision time estimate for grading. When the perceived confidence is low, i.e., the estimated standard deviation is large, the corrected collision time is reduced accordingly, and the braking triggering timing is automatically advanced, realizing a conservative braking strategy in low-confidence scenarios to ensure safety.

[0045] Preferably, the first set multiple is 2, that is: ; The safety correction collision time The first threshold calculated in step S3 Second threshold Compare them.

[0046] like This indicates that the current collision risk is low, no intervention is needed, braking is not triggered, and the system returns to step S1 to reacquire data and enter the next monitoring cycle.

[0047] like This indicates that the collision risk has reached the point where intervention is required, but there is still ample reaction time. At this point, the smooth braking mode is activated and step S5 is executed.

[0048] like This indicates that the risk of collision is extremely urgent and braking at maximum capacity is necessary immediately. At this point, the emergency braking mode is activated and step S6 is executed.

[0049] The exit braking judgment uses a more lenient approach to prevent frequent mode switching, and incorporates a hysteresis anti-shake mechanism. The entry braking judgment uses a conservative value. Mode switching requires confirmation over multiple consecutive frames to take effect, preventing accidental triggering due to transient interference. Emergency braking mode has the highest priority; once triggered, it cannot immediately revert to smooth mode but must recover to normal mode after a collision time. Exit step by step.

[0050] In the hierarchical comparison, redundancy checks and boundary condition handling are also introduced. When There is no risk of collision, and the collision time is set to infinity. Fusion distance. As a redundancy check, emergency braking is forcibly triggered when the distance is extremely close. Obstacle azimuth angle. Used to determine the direction of obstacle intrusion in non-directly facing scenes and adjust the braking attitude plane allocation. Sensor health flag bit. Automatically expand equivalent value when sensor degrades. Decision-making has become more conservative.

[0051] In step S5, under smooth braking mode, the braking stiffness coefficient is determined based on the current total mass of the UAV. A continuously increasing deceleration command is generated according to the power function law, using the braking stiffness coefficient as a parameter and the difference between the second threshold and the current collision time estimate as the time margin. The deceleration command value when it drops to the first threshold is used as the reference trajectory. First, based on the current total mass of the drone Determine the braking stiffness coefficient The braking stiffness coefficient describes the response stiffness of the braking system, that is, the rate of deceleration increase per unit time margin, and its physical meaning is similar to that of spring stiffness. A larger value results in a stiffer braking response, while a smaller value results in a softer response; essentially, it's an adaptive adjustment factor for the braking gain. The braking stiffness coefficient... With the current total mass of the drone They satisfy a linear relationship ,in, Based on the basic stiffness coefficient, Both are mass impact coefficients, obtained through calibration experiments on the target UAV platform, and are applicable to weights ranging from 12kg to 30kg.

[0052] Then, with the second threshold Compared with the current collision time estimate difference As a remaining time margin, this margin reflects the time leeway before emergency braking must be initiated. A normalized time constant is introduced. The value is set to 1 second, and the time margin is dimensionless to make the stiffness coefficient... The physical unit is constant in m / s², and is not affected by the smoothing factor. The value selection has an impact, which facilitates parameter reuse and calibration migration under different system architectures and sampling frequencies.

[0053] The time margin is mapped to a deceleration command according to the power function principle, specifically as follows: ; in, This indicates a deceleration command, with units of _____. The lower limit is 0, so no braking command is generated; the upper limit is the maximum available deceleration. If the limit is exceeded, the amplitude will be limited; the typical working range is 0 to 0.85× ; Indicates the braking stiffness coefficient; Indicates the second threshold; This represents the estimated collision time. This represents the preset normalized time constant; This represents the smoothing factor, with a value ranging from 1 to 2.

[0054] Smoothing factor When the value is greater than 1, the deceleration command increases slowly in the initial stage of braking and accelerates at the end of braking, exhibiting a characteristic of initial gentleness followed by stiffness. The braking is a double parabolic braking system. The initial braking is extremely smooth to protect the fuselage and load, with minimal attitude disturbance. As the collision time decreases, the deceleration and acceleration increase to ensure safety.

[0055] During smooth braking, the deceleration command increases continuously from zero, and the estimated collision time correspondingly increases from the second threshold. To the first threshold Directional movement. When the collision time estimate... Drop to the first threshold At this time, the current deceleration command value is saved as a reference trajectory and passed to step S6 as the tracking target for the emergency braking mode. The deceleration command has an upper limit on its rate of change. This is used to prevent extreme jumps and further ensure smoothness.

[0056] In step S6, under the emergency braking mode, the optimal deceleration command is solved by constraint optimization with the reference trajectory as the target. When safety corrects collision time Not greater than the first threshold At this point, the system enters emergency braking mode. In this mode, the reference trajectory transmitted in step S5 is used as the deceleration tracking target. Switching point, power function output value for smooth braking Aligned with the initial objective of constraint optimization, the switching is seamless and without any sudden changes or shocks.

[0057] The specific process of constraint optimization is as follows: Using the reference trajectory as the deceleration tracking target, a multi-objective cost function is constructed, including deceleration tracking error, attitude angle deviation penalty, and thrust rate of change penalty. The deceleration tracking error is used to track the reference trajectory, the attitude angle deviation penalty is used to ensure attitude safety, and the thrust rate of change penalty is used to ensure smoothness. The upper and lower limits of the UAV motor speed, the upper limit of the pitch angle, and the upper limit of the thrust rate of change are used as constraints. Within each control cycle, minimizing the cost function is the optimization objective. The optimal control sequence within a preset time domain is solved within the feasible region defined by the constraints, and the first control variable in the optimal control sequence is output as the optimal deceleration command for the current cycle.

[0058] At this point, the reference deceleration has usually reached or is close to the maximum available deceleration. The core function of constraint optimization is to rationally decompose the total braking thrust demand to each rotor, coordinate the pitch and roll attitude distribution, achieve the fastest deceleration under the premise of strictly adhering to physical constraints, and prevent attitude loss of control or power saturation caused by improper thrust distribution.

[0059] Preferably, the constraint optimization is achieved using model predictive control. The motor speed limit includes an upper and lower limit for each motor speed, the pitch angle limit is the maximum permissible pitch angle of the UAV, and the thrust rate of change limit is the maximum permissible change in thrust within a single control cycle.

[0060] In step S7, the optimal deceleration command is converted into pitch angle control and total thrust control, and output to the flight control system to perform braking. At the same time, the current operating status data of the UAV is acquired and returned to participate in the state estimation and closed-loop correction of the next cycle.

[0061] Convert the optimal deceleration command output in step S5 or S6 into a pitch angle command for the UAV. and total thrust command For multi-rotor UAVs, deceleration commands are achieved by tilting the fuselage to generate a horizontal thrust component. The required pitch angle is calculated based on the magnitude of the deceleration, and altitude stability is maintained through total thrust control. The pitch angle control and total thrust control are output to the attitude control loop of the flight control system, updated at a frequency of 200Hz and aligned with the inner loop of the flight control system. The flight control system then distributes these commands to the speed commands of each motor, driving the motors to perform braking.

[0062] Simultaneously, the inertial measurement unit provides real-time feedback on the aircraft's acceleration and angular velocity data. This high-frequency data is also sent to the flight control attitude loop and fusion module for time alignment interpolation, serving as part of the UAV's current operational state data, and is returned to participate in the state estimate for the next cycle. This completes the full closed-loop control cycle of perception, decision-making, execution, and feedback, and the next cycle restarts from step S1.

[0063] During the braking exit phase, when the relative velocity of the obstacle is detected in multiple consecutive frames... And fusion distance When the deceleration continues to increase, the collision risk is determined to be eliminated. The deceleration command exits in an exponential decay manner, with the deceleration smoothly decreasing from the current value along an exponential curve to zero, avoiding sudden release that could cause attitude rebound. The specific exponential decay law is as follows: ; in, Indicates the current moment; Indicates the start time of exit; This indicates the deceleration command at the current moment; The deceleration command indicating exit from the start time. Indicates the current moment; This represents the decay time constant, in seconds. The pitch angle control value synchronously returns to zero according to an exponential law until the braking state is completely disengaged and normal flight is restored.

[0064] Thus, the adaptive smooth obstacle avoidance braking method for medium and large-sized UAVs provided in Embodiment 1 of the present invention systematically solves the core technical problem in obstacle avoidance braking of medium and large-sized multi-rotor UAVs through the complete closed-loop process of steps S1 to S7.

[0065] This method uses collision time as a unified decision factor, fundamentally replacing the traditional fixed distance threshold scheme and ensuring a strict match between braking trigger timing and actual braking requirements. By introducing real-time UAV mass parameters to construct an adaptive braking stiffness model, automatic adaptation across the entire load range is achieved without manual parameter readjustment. A power-law deceleration generation curve is employed to eliminate step response at the control input source, effectively suppressing attitude overshoot and structural impact. Combined with an error-state extended Kalman filter multi-source fusion sensing framework, it fully leverages the complementary advantages of LiDAR, visual cameras, and millimeter-wave radar, while also possessing smooth degradation capabilities in the event of sensor failure, ensuring system reliability under complex sensing conditions.

[0066] The aforementioned technical features work synergistically. The mass adaptive model simultaneously serves threshold calculation and stiffness determination; power function smooth braking and constraint-optimized emergency braking seamlessly connect at the switching point; and safety-corrected collision time integrates perception confidence into the decision-making process. These technical features collectively achieve a comprehensive technical effect of full-load adaptive braking, low-impact smooth braking, and high-fault-tolerant perception, providing reliable technical assurance for the safe flight of industrial-grade UAVs in complex environments.

[0067] Example 2 To fully illustrate the implementation process of the adaptive smooth obstacle avoidance and braking method for medium and large UAVs proposed in Embodiment 1 of this invention, Embodiment 2 of this invention provides a detailed process for heavy-load conditions.

[0068] 1. Simulation / Experimental Environment: This embodiment 2 uses a custom-designed hexacopter drone as the test platform. The drone has a payload range of 12kg to 25kg, a wheelbase of 1200mm, and is powered by a T-Motor U8 Lite propeller with 30-inch carbon fiber rotors. The drone's current total mass... Under moderate to heavy load conditions. Initial flight speed The obstacle is a stationary column with a diameter of 1.5m and an initial distance of... Maximum available deceleration This value already includes the equivalent effect of aerodynamic drag. Total system delay. Environmental conditions were calm with good visibility. Normalized time base. .

[0069] 2. Calculation process: First, determine the braking stiffness coefficient based on the current total mass of the UAV. , and They satisfy a linear relationship Substituting the preferred parameters into this embodiment , as well as The calculation yields:

[0070] Then, the two-stage braking thresholds are calculated. The pure deceleration time is the ratio of the current flight speed to the maximum available deceleration: ; Second threshold This is the sum of the pure deceleration time and the total system delay: ; First threshold The sum of half the pure deceleration time and the total system delay: ; Initial collision time The ratio of initial distance to relative velocity: ; because ,Right now The system is in normal flight mode and does not trigger braking.

[0071] 3. Simulation of braking process: As the drone continued to approach the obstacle... Gradually decreasing. When At 3.3 seconds, the second threshold is reached, and the system initiates smooth braking. The deceleration increases according to a power function: ; because The deceleration exhibits a quadratic parabolic growth characteristic, initially gentle and then stiff. In the initial stage of braking, the deceleration increases slowly with minimal attitude disturbance; as... The rate of decrease continues to shorten, and the rate of increase in the rate of decrease is accelerating.

[0072] when When the speed drops to 1.8 seconds, the first threshold is reached, and the deceleration command value at this time is: ; This value has exceeded the maximum available deceleration. The system triggers a limit and uses this value as a reference trajectory, simultaneously entering emergency braking mode. In emergency braking mode, the MPC uses this reference trajectory as the target and performs optimization solutions under constraints of motor speed limits, pitch angle limits, and thrust rate of change limits, rationally decomposing the total braking thrust demand to each rotor and coordinating pitch and roll attitude distribution. The MPC solves for the optimal control sequence within a finite time domain in each control cycle, executing the first step to ensure that the actual deceleration tracks the reference trajectory and does not violate physical constraints.

[0073] Ultimately, the drone came to a safe stop about 5.3 meters from the obstacle. Throughout the process, the drone's pitch angle was kept within the optimal performance range of 12° to 15°, and the gimbal image blur was reduced by 80%.

[0074] 4. Key parameter curve characteristics: The curve shows a linear downward trend during the constant velocity phase. After braking is initiated, the drone begins to decelerate. The descent rate gradually slowed down, and when it finally came to a stop, the remaining distance was about 5.3m.

[0075] The curve starts from zero and grows according to a parabolic law. After reaching its peak, the point approaches... And maintain this state, with no step transitions throughout.

[0076] The attitude angle curve increases smoothly in an approximately S-shape, with the maximum pitch angle being... The 12.8° angle shows no overshoot or oscillation, indicating that the attitude response has good damping characteristics.

[0077] The Jerk curve grows linearly from zero to approximately 10.5 during the smoothing braking phase. There were no sudden spikes throughout the entire process.

[0078] This embodiment 2 is quantitatively compared with the comparative scheme. The comparative scheme uses the default position loop parameters of PX4 v1.13 firmware, sets a fixed distance threshold of 10 meters to trigger emergency stop, and is measured under the same platform and working conditions.

[0079] Table 1: Quantitative Comparison Table between Example 2 and the Comparative Scheme

[0080] The comparison results show that, under heavy load conditions, the pitch angle fluctuation of the present invention is reduced from 15.2° to 12.8°, a reduction of 15.8%; the Jerk peak value is reduced from 38.6 m / s³ to 11.2 m / s³, a reduction of 71.0%; the braking end distance is increased from 4.8 m to 5.3 m, with a safety margin improvement of 10.4%; the gimbal image blur is reduced from 100% to 20%, a reduction of 80%; and the attitude overshoot is completely eliminated from 22% to 0%. All indicators are significantly better than the comparative scheme, verifying the effectiveness of the present invention under heavy load conditions.

[0081] Example 3 To fully illustrate the implementation process of the adaptive smooth obstacle avoidance and braking method for medium and large UAVs proposed in Embodiment 1 of this invention, Embodiment 3 of this invention provides a detailed process for light-load conditions.

[0082] 1. Simulation / Experimental Environment: Example 3 uses a custom-designed hexacopter drone as the test platform. This drone has a payload range of 12kg to 25kg, a wheelbase of 1200mm, and is powered by a T-Motor U8 Lite propeller with 30-inch carbon fiber rotors. The drone's current total mass... It is in a light load condition. Initial flight speed The obstacle is a stationary column with a diameter of 1.5m and an initial distance of... Maximum available deceleration Total system latency Normalized time base The environmental conditions are calm with good visibility.

[0083] 2. Calculation process: First, determine the braking stiffness coefficient based on the current total mass of the UAV. Substitute And the preferred parameters of this embodiment , The calculation yields: ; The stiffness coefficient is lower than 1.94 m / s² in the heavy-load condition of Example 1, which reflects the softer braking response under light load, which is beneficial to ensuring smoothness.

[0084] Then calculate the two-stage braking thresholds. The pure deceleration time is: ; Second threshold This is the sum of the pure deceleration time and the total system delay: ; First threshold The sum of half the pure deceleration time and the total system delay: ; Initial collision time The ratio of initial distance to relative velocity: ; because ,Right now The system is in normal flight mode and does not trigger braking.

[0085] 3. Simulation of braking process: As the drone continued to approach the obstacle... As the deceleration gradually decreases to 2.25 seconds, it reaches the second threshold, and the system initiates smooth braking. The deceleration increases according to a power function: ; Due to its smaller stiffness coefficient under light load, the deceleration rate increases more slowly, resulting in a smoother initial braking and effectively protecting the fuselage structure and precision loads.

[0086] As braking continues, the deceleration increases continuously from zero. Continue descending towards the first threshold. When it drops to 1.41s, the deceleration command value reaches approximately The system saves the deceleration command value at this moment as a reference trajectory and enters emergency braking mode. In emergency braking mode, the MPC uses this reference trajectory as a target and solves for the optimal deceleration command under constraints, ensuring that the actual deceleration tracks the reference trajectory without exceeding physical limits. During the emergency braking phase, the reference deceleration rapidly increases and approaches the maximum available deceleration. MPC rationally distributes the total braking thrust to each rotor, coordinating pitch and roll attitude distribution to achieve the fastest deceleration.

[0087] The drone eventually came to a safe stop about 4.8 meters from the obstacle.

[0088] Verification results: This embodiment 3 verifies the good adaptability of the present invention under light load conditions. Compared with the heavy load condition in Embodiment 1, the stiffness coefficient automatically decreases under light load, the braking response is softer, the maximum pitch angle decreases from 12.8° to 11.5°, and the Jerk peak value decreases from 11.2 m / s³ to 8.7 m / s³, resulting in a smoother braking process. Simultaneously, the final braking distance is 4.8 m, meeting safety requirements. The braking distance consistency under light load reaches ±0.2 m, significantly better than the ±0.8 m of the comparative scheme, verifying the technical advantages of the present invention's full-load adaptive design.

[0089] 4. Key parameter curve characteristics: The curve descends at a constant speed starting at 4s. After braking is triggered at 2.52s, the descent rate gradually slows down, and the remaining distance when it finally comes to a stop is about 4.8m. The curve starts from zero and grows according to a parabolic law, approaching the maximum available deceleration after entering the emergency zone. And maintain it. The attitude angle curve increases smoothly in an approximately S-shape, with the maximum pitch angle approximately... 11.5°, no overshoot and no oscillation. The Jerk curve is smooth throughout, with a peak value of approximately [missing value]. It performs better under light load conditions than under heavy load conditions.

[0090] This embodiment 3 is quantitatively compared with the comparative scheme. The comparative scheme uses the default position loop parameters of PX4 v1.13 firmware, sets a fixed distance threshold of 10 meters to trigger emergency stop, and is measured under the same platform and working conditions.

[0091] Table 2: Quantitative Comparison Table of Example 3 and Comparative Scheme

[0092] The comparison results show that the present invention reduces pitch angle fluctuation by 16.7% and Jerk peak value by 75.3% under light load conditions, increases braking end distance safety margin by 23.1%, and improves braking distance consistency accuracy by 75%. All indicators are significantly better than the comparison scheme.

[0093] Example 4: High-speed moving obstacle scenario 1. Simulation / Experimental Environment: This embodiment 4 uses the same drone platform as embodiment 2. Initial flight speed Flying horizontally forward. The obstacle is an oncoming ground vehicle, speed... Moving towards each other, initial relative distance Relatively close speed Maximum available deceleration In the sensor configuration, millimeter-wave radar is the primary sensor for velocity measurement, while lidar is used for ranging.

[0094] 2. Calculation process: Millimeter-wave radar directly observes the radial velocity of obstacles, which is then confirmed by ES-EKF fusion to determine the relative velocity. Two threshold levels are calculated based on the drone's own braking capability. The pure deceleration time is: ; Second threshold This is the sum of the pure deceleration time and the total system delay: ; First threshold The sum of half the pure deceleration time and the total system delay: ; Initial collision time The ratio of initial distance to relative velocity: ; stiffness coefficient The same as the heavy-load condition in Example 2 3. Simulation of braking process: Due to the high relative speed . The descent rate was extremely rapid. Approximately 3.9 seconds later, The deceleration decreased from 5.22s to 2.3s, triggering smooth braking. The deceleration increased rapidly according to a power function.

[0095] The rate of ascent is significantly faster than in a stationary obstacle scenario. Within approximately 1 second, That is, the deceleration drops to 1.3 seconds, at which point the system enters emergency braking mode. In emergency braking mode, MPC performs constraint optimization with the reference trajectory as the target. At this point, the reference deceleration is close to the maximum available deceleration. The MPC (Multi-Purpose Control) system rationally distributes the total braking thrust to each rotor, coordinating pitch and roll attitude distribution to achieve the fastest deceleration while strictly adhering to motor speed limits, pitch angle limits, and thrust change rate limits. Ultimately, the UAV safely came to a stop approximately 4.5 meters from the obstacle, with the pitch angle smoothly increasing throughout the process. 12.5°, no overshoot.

[0096] 4. Key parameter curve characteristics: The descent rate is fast, but after braking is initiated, the descent rate gradually slows down as the drone decelerates. The curve's ascent rate is significantly faster than in a stationary obstacle scenario, entering the emergency zone in approximately 1 second. The attitude angle curve increases rapidly but smoothly to... 12.5°. The peak value of the Jerk curve is approximately... It is slightly higher than that of a static scene, but there are no sudden spikes throughout.

[0097] This embodiment 4 is quantitatively compared with the comparative scheme. The comparative scheme uses a fixed distance threshold to trigger braking, and is measured under the same platform and working conditions.

[0098] Table 3: Quantitative Comparison Table between Example 4 and the Comparative Scheme

[0099] The comparison results show that the proposed solution combines millimeter-wave radar velocity observation with ES-EKF fusion. The estimation error was reduced from ±0.8s to ±0.15s, an 81% improvement in accuracy. The minimum safe distance was increased from 1.2m to 4.5m, a 275% increase in safety margin. Pitch angle fluctuation was reduced by 25.6%, and Jerk peak was reduced by 72.4%. These results verify that the present invention can accurately perceive collision risks and trigger braking in a timely manner in high-speed moving obstacle scenarios, demonstrating sufficient dynamic response capabilities.

[0100] Example 5: Scenario where the obstacle is not directly in front of the scene 1. Simulation / Experimental Environment: This embodiment 5 uses the same drone platform as embodiment 2. , It flies along the positive X-axis. The obstacle is a moving target intruding from the right side, initially offset 15m in the Y direction and 40m away in the X direction. The obstacle's speed... The vehicle moves along its Y-axis trajectory. The initial obstacle azimuth angle is approximately 20.6°, located to the right and slightly ahead. Predicted minimum encounter distance. This indicates that without intervention, the obstacle will be encountered at the drone's waypoint. Maximum available deceleration. Maximum lateral acceleration .

[0101] 2. Calculation process: ES-EKF fusion outputs obstacle trajectory and predicts nearest encounter time. , equivalent Equivalent relative velocity Vertical second threshold for: ; A combined braking strategy is adopted, and the deceleration is decomposed in the opposite direction of the relative velocity: ; ; in, It is the angle between the direction of the relative velocity of the obstacle and the flight direction of the drone.

[0102] 3. Simulation of braking process: Upon detecting a lateral intrusion risk, the system employs a combined braking strategy of longitudinal deceleration and lateral fine-tuning. Longitudinal deceleration is primary, used to reduce the drone's forward speed and increase safety margin. Lateral deceleration is secondary, used to fine-tune the drone's lateral position to avoid obstacle intrusion paths. Dual-axis coordinated braking prevents coupled oscillations.

[0103] As obstacles cross the drone's flight path, the relative azimuth gradually changes from the front right to the right and then to the rear right. First decrease, then increase. Smooth braking is triggered at 2.8 seconds, with deceleration increasing according to a power function. Once the obstacle has completely passed the flight path, the relative distance begins to increase. As the system recovers, it enters the braking disengagement process. The deceleration command is released smoothly in an exponential decay manner, and the pitch and roll angles return to zero simultaneously.

[0104] 4. Key parameter curve characteristics: The curve first decreases and then increases, reflecting the process of the obstacle moving away after it crosses the path. The curve is dominated by the longitudinal component, with a smaller lateral component, and is used for fine-tuning avoidance. The pitch angle is the largest component in the attitude angle curve. 10°, with a maximum roll angle of approximately 5°, and smooth variation along both axes. The synthesized Jerk peak is approximately... Uncoupled oscillation.

[0105] This embodiment 5 is quantitatively compared with the comparative scheme, which adopts a single-axis emergency stop strategy, only performing longitudinal deceleration without lateral adjustment.

[0106] Table 4: Quantitative Comparison Table between Example 5 and the Comparative Scheme

[0107] The comparison results show that the proposed solution, through a combined braking strategy, increases the minimum lateral clearance from 2.1m to 5.8m, improving the safety margin by 176%. The maximum roll angle fluctuation is reduced by 59.3%, the yaw angle deviation by 75.3%, and the time to recover the original trajectory is shortened by 54.8%. These results verify that the proposed solution can accurately determine the direction of obstacle intrusion and rationally distribute braking force across axes in scenarios where the obstacle is not directly facing it, achieving coordinated and smooth braking across multiple axes.

[0108] Example 6: Sensor Failure Degradation Mode 1. Simulation / Experimental Environment: This embodiment 6 uses the same drone platform as embodiment 2. , The obstacle is a stationary column, with an initial distance of... The fault setting is: fly to In the event of simultaneous failure of both the lidar and millimeter-wave radar, simulating an electromagnetic interference scenario, the system degrades to a binocular vision-only operating mode. The binocular camera has a baseline of 150mm, an effective ranging range of 0.5m to 30m, and a ranging error model of [missing information]. The error is small at close range, but increases quadratically at long range. After downgrading... Dynamically adjusted based on the visual ranging error model The greater the distance, the lower the confidence level.

[0109] 2. Calculation process: Before the fault, the three sensors were fused together. , The estimation accuracy is high. When a fault occurs, the ES-EKF detects a continuous exceedance of the innovation limits of both the lidar and millimeter-wave radar using a chi-square test of the innovation sequence, determining that the sensor has failed. Flag position 0. The observation weights of the faulty sensors are continuously and smoothly reduced to zero, while the fusion algorithm architecture remains unchanged, without the need to switch modes.

[0110] After the downgrade, only visual fusion is performed. Dynamically calculated using a visual ranging error model. When the distance is 30m... When the distance is 20m, Safety margin As distance decreases The safety margin is adaptively narrowed as the distance decreases. The braking trigger timing is determined by distance and... The decision is made jointly: more conservative at long distances, more precise at close distances.

[0111] 3. Downgrade process: The moment of failure It is estimated that only a slight fluctuation occurred, less than 0.3 seconds, after which the visual fusion tracking transitioned smoothly without interruption. This was due to the long distance. Increase The deceleration is reduced accordingly, and braking trigger is slightly advanced, ensuring safety redundancy after degradation. Throughout the degradation process, the deceleration curve maintains its power function characteristics, the smoothness of the attitude angles is unaffected, and the maximum pitch angle is approximately... 12.2°.

[0112] 4. Key parameter curve characteristics: Sensor health flags are in The health status of the lidar and millimeter-wave radar changes from 1 to 0. The curve increases immediately after a fault, and decreases quadratically as the distance decreases. The curve experienced a momentary, slight fluctuation after the fault, followed by a smooth transition without interruption. Curve braking triggers slightly earlier, and the curve shape maintains its power function characteristics. The smoothness of the attitude angle curve remains unaffected before and after degradation, with a maximum of approximately 12.2°.

[0113] In this embodiment 6, a quantitative comparison with the comparative scheme shows that the overall system of the comparative scheme is severely degraded after the radar fails, resulting in braking interruption or false triggering.

[0114] Table 5: Quantitative Comparison Table between Example 6 and the Comparative Scheme

[0115] The comparison results show that the present invention achieves a shockless transition when the sensor fails through the ES-EKF information detection and smooth degradation mechanism. The estimated interruption time has been reduced from over 1.5 seconds to less than 0.1 seconds, a reduction of over 93%. The minimum safe distance has been increased from 2.8 meters to 5.1 meters, improving the safety margin by 82%. These results verify that even in the event of sensor failure, the binocular vision system alone can still perform safe braking, meeting the requirements for fault operation.

[0116] Example 7: Scenario where obstacles disappear or deviate during braking. 1. Simulation / Experimental Environment: This embodiment 7 uses the same drone platform as embodiment 2. , The initial scenario involves an obstacle in front. The deceleration has decreased to 2.5 seconds; the system is in the smooth braking phase, and the current deceleration rate is... The event is set to obstacle deviation, such as a vehicle turning out of its track, resulting in a relative speed... The value changes from positive to negative. The disappearance criterion is that the value is detected in multiple consecutive frames. and The rate of increase continues. The braking exit strategy is an exponentially decaying release with a time constant. .

[0117] 2. Calculation process: ES-EKF tracking The speed gradually decreases from +12 m / s, becoming negative after crossing zero, indicating that the relative distance between the obstacle and the drone begins to increase. When The risk of collision has been eliminated. Set to infinity. The exit condition is... ,in, This is a hysteresis band, and multiple consecutive frames of confirmation are required before exiting. Deceleration is released according to an exponential decay law. ; in, The decay time constant is 0.5s. The pitch angle synchronously returns to zero degrees according to an exponential law to avoid sudden pitching that could cause a sudden change in lift.

[0118] 3. Exit process: Upon detecting an obstacle deviation, the deceleration does not jump from its current value to zero, but rather decays smoothly along an exponential curve. After approximately three times the time constant, or 1.5 seconds, the deceleration is essentially complete. The pitch angle recovers smoothly and synchronously, without reverse overshoot, the fuselage exhibits no nose-diving, and altitude fluctuations are kept within a small range. Throughout the exit process, Jerk is negative, with an absolute value of approximately [missing value]. The entire process is smooth and shock-free.

[0119] 4. Key parameter curve characteristics: The curve increases rapidly and becomes safe after the obstacle is deviated from. The curve decays exponentially to zero without any abrupt changes. The attitude angle curve from... The temperature smoothly rises from approximately 8° to 0° without reverse overshoot. The Jerk curve is negative during the exit phase, with an absolute value of approximately... The entire process is smooth.

[0120] This embodiment 7 is quantitatively compared with the comparative scheme. The comparative scheme adopts a switch-type release strategy, which releases all braking force immediately after the obstacle deviates.

[0121] Table 6: Quantitative Comparison Table between Example 7 and the Comparative Scheme

[0122] The comparison results show that the present invention uses an exponential decay exit mechanism, and Jerk exits from [a certain point] at the moment of exit. Reduce to The following reductions were achieved: Attitude overshoot decreased from +6° to within 0.5°, a reduction of 92%; Altitude fluctuation decreased from ±1.2m to ±0.2m, a reduction of 83%; and Time to regain stable flight was shortened from 3.5s to 1.2s, a reduction of 65.7%. These results demonstrate that the present invention can achieve smooth exit in scenarios where obstacles disappear or deviate from the target, avoiding attitude rebound and impact caused by sudden release of braking force.

[0123] Example 8: Robust braking scenario under strong wind interference 1. Simulation / Experimental Environment: This embodiment 8 uses the same drone platform as embodiment 2. , The obstacle is a stationary column, with an initial distance of... The wind conditions are a stable headwind of 6 m / s superimposed with a gust of 1.5 m / s, where the gust exhibits a sinusoidal variation. The maximum available deceleration... This is the set value for windless conditions. In windy conditions, aerodynamic drag contributes additionally to braking deceleration. The wind-resistant speed loop is activated in the flight control inner loop, while the outer loop braking algorithm remains unchanged. This test aims to verify the robustness of the algorithm; the changes in aerodynamic characteristics with airspeed have been verified through the attitude control robustness of the flight control inner loop. Safety margins are provided, and the braking algorithm itself does not rely on an accurate aerodynamic model.

[0124] 2. Calculation process Calculated based on ground speed and relative speed to obstacles The threshold remains as nominal. Conservative calculations, with a safety margin: Second threshold: ; stiffness coefficient Determined by mass, not directly affected by wind. Gusts cause... Slight fluctuations are expected, and the hysteresis mechanism, including multi-frame confirmation and hysteresis bands, effectively suppresses mode jitter. Quality estimates are paused during gusts, and estimates from the pre-wind constant velocity period are used to avoid contaminating the quality estimates with wind disturbance acceleration.

[0125] 3. Simulation of braking process: Under headwind conditions, airspeed increases, and aerodynamic drag contributes an additional approximately The actual braking distance is slightly shorter than under calm conditions due to deceleration. Gusts of wind cause this. and There were slight fluctuations, but the smoothness of the power function curve and the upper limit of its rate of change remained. The combined effect of amplitude limiting suppresses the deceleration fluctuations caused by gusts within a small range. Pitch angle fluctuations are controlled within ±1.2°, with no oscillations or divergence.

[0126] 4. Key parameter curve characteristics: The curve shows an overall downward trend, with slight fluctuations, and the fluctuation range caused by gusts is about ±0.15s. The curve is a power function with a clear main trend, superimposed with tiny ripples, and the peak value is approximately... Not triggered Limiting. Maximum approximation of attitude angle curve. 12.2°, accompanied by gusts of wind up to ±1.2°, with no overshoot or divergence. The peak value of the Jerk curve is approximately... It is slightly higher than that under windless conditions, but still far lower than that of the comparison scheme.

[0127] This embodiment 8 is quantitatively compared with the comparative scheme. The comparative scheme uses a fixed threshold to trigger braking and is measured under the same platform and working conditions.

[0128] Table 7: Quantitative Comparison Table between Example 8 and the Comparative Scheme

[0129] The comparative results show that, under strong wind interference, the maximum pitch angle fluctuation of the proposed solution is reduced from 18.5° to 13.4°, a reduction of 27.6%. The Jerk peak value is reduced from 52.1 m / s³ to 11.8 m / s³, a reduction of 77.4%. The braking distance is increased from 3.2 m to 5.6 m, improving the safety margin by 75%. Attitude oscillations are eliminated completely, exhibiting monotonically converging, instead of the previous 3-4 decaying oscillations. Altitude fluctuations are reduced from ±2.1 m to ±0.5 m, a reduction of 76.2%. These results verify that the proposed solution maintains attitude stability under strong wind interference, without oscillation divergence, and that the inherent low-pass filtering characteristics of the power function curve have a significant suppressive effect on wind disturbance.

[0130] Example 9 Based on the adaptive smooth obstacle avoidance braking method for medium and large-sized UAVs proposed in Embodiment 1 of the present invention, Embodiment 9 of the present invention also proposes an adaptive smooth obstacle avoidance braking device for medium and large-sized UAVs. Figure 2 This is a schematic diagram of an adaptive smooth obstacle avoidance braking device for medium and large-sized UAVs proposed in Embodiment 9 of the present invention.

[0131] At the hardware level, electronic device 200 includes a processor 210, and optionally, an internal bus 220, a network interface 230, and memory. The memory may include main memory, such as high-speed random-access memory (RAM), or it may also include non-volatile memory, such as at least one disk drive. Of course, the electronic device may also include other hardware required for other business operations.

[0132] The processor 210, network interface 230, and memory can be interconnected via an internal bus 220. This internal bus 220 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be categorized as an address bus, data bus, control bus, etc. For ease of illustration, only a single bidirectional arrow is used in this diagram, but this does not imply that there is only one bus or one type of bus. The memory is used to store programs. Specifically, the program can include program code, which includes computer operation instructions. The memory can include main memory 240 and non-volatile memory 250, and provides instructions and data to the processor 210.

[0133] Processor 210 reads the corresponding computer program from non-volatile memory 250 into memory 240 and runs it, forming a host platform context measurement and extended remote verification device at the logical level. Processor 210 executes the program stored in memory and specifically performs the following: In step S1, obstacle observation data and UAV operating status data are acquired.

[0134] In step S2, after synchronizing the acquired obstacle observation data and UAV operating status data, the synchronized UAV operating status data is used as a known reference quantity. The relative motion state of the obstacle is estimated by fusing the data through a state estimation filtering algorithm. The collision time estimate and the estimated standard deviation of the collision time estimate are calculated based on the relative motion state. In step S3, the maximum available deceleration is determined based on the current total mass of the UAV. A first threshold for triggering emergency braking and a second threshold for triggering smooth braking are calculated based on the current flight speed, the maximum available deceleration, and the total delay from detection to braking. The first threshold is less than the second threshold. In step S4, a safe corrected collision time is determined based on the estimated collision time and the estimated standard deviation. The safe corrected collision time is compared with a first threshold and a second threshold: if the safe corrected collision time is greater than the second threshold, braking is not triggered and the system returns to reacquire obstacle observation data and UAV operational status data; if the safe corrected collision time is between the first threshold and the second threshold, the system enters a smooth braking mode and executes step S5; if the safe corrected collision time is not greater than the first threshold, the system enters an emergency braking mode and executes step S6. In step S5, under smooth braking mode, the braking stiffness coefficient is determined based on the current total mass of the UAV. A continuously increasing deceleration command is generated according to the power function law, using the braking stiffness coefficient as a parameter and the difference between the second threshold and the current collision time estimate as the time margin. The deceleration command value when it drops to the first threshold is used as the reference trajectory. In step S6, under the emergency braking mode, the optimal deceleration command is solved by constraint optimization with the reference trajectory as the target. In step S7, the optimal deceleration command is converted into pitch angle control and total thrust control, and output to the flight control system to perform braking. At the same time, the current operating status data of the UAV is acquired and returned to participate in the state estimation and closed-loop correction of the next cycle.

[0135] Figure 1It can be applied to processor 210, or implemented by processor 210. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit in the hardware of the processor or by instructions in the form of software. The processor mentioned above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this application can be directly embodied in the execution of the hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0136] Of course, in addition to software implementation, the electronic device of this application does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. In other words, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0137] The description of the relevant parts of the adaptive smooth obstacle avoidance braking device for medium and large UAVs provided in Embodiment 2 of this application can be found in the detailed description of the corresponding parts of the adaptive smooth obstacle avoidance braking method for medium and large UAVs provided in Embodiment 1 of this application, and will not be repeated here.

[0138] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that the elements inherent in a process, method, article, or apparatus that includes a list of elements are included. 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 said element. Additionally, portions of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0139] While specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art can make other modifications or variations based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. An adaptive smooth obstacle avoidance and braking method for medium and large-sized unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: After synchronizing the acquired obstacle observation data and UAV operating status data, the synchronized UAV operating status data is used as a known reference quantity. The relative motion state of the obstacle is estimated by fusing the data through a state estimation filtering algorithm. The collision time estimate and the estimated standard deviation of the collision time estimate are calculated based on the relative motion state. The maximum available deceleration is determined based on the current total mass of the UAV. A first threshold for triggering emergency braking and a second threshold for triggering smooth braking are calculated based on the current flight speed, the maximum available deceleration, and the total delay from detection to braking. The first threshold is less than the second threshold; Based on the estimated collision time and the estimated standard deviation, a safety correction collision time is determined. The safety correction collision time is compared with a first threshold and a second threshold: if the safety correction collision time is greater than the second threshold, braking is not triggered and the system returns to reacquire obstacle observation data and UAV operational status data; if the safety correction collision time is between the first threshold and the second threshold, a smooth braking mode is entered; if the safety correction collision time is not greater than the first threshold, an emergency braking mode is entered. In smooth braking mode, the braking stiffness coefficient is determined based on the current total mass of the UAV. According to the power function law, the braking stiffness coefficient is used as a parameter and the difference between the second threshold and the current collision time estimate is used as the time margin to generate a continuously increasing deceleration command. The deceleration command value when it drops to the first threshold is used as the reference trajectory. In the emergency braking mode, the optimal deceleration command is solved by constraint optimization with the reference trajectory as the target. The optimal deceleration command is converted into pitch angle control and total thrust control, and output to the flight control system to perform braking. At the same time, the current operating status data of the UAV is acquired and returned to participate in the state estimation and closed-loop correction of the next cycle.

2. The method according to claim 1, characterized in that, The detailed process of synchronizing the acquired obstacle observation data and UAV operational status data is as follows: Obstacle observation data acquired: obstacle distance and azimuth angle are collected by airborne lidar; depth map of obstacles is collected by airborne visual camera; radial velocity data of obstacles is collected by airborne millimeter-wave radar. Acquire UAV operational status data: The flight control system outputs the UAV's current position, speed, and attitude in real time; the inertial measurement unit collects the three-axis acceleration and three-axis angular velocity of the UAV in real time. The obstacle observation data and the UAV operation status data are each attached with a timestamp in the form of data frames. Based on the timestamps of each data, the data with different sampling frequencies are interpolated and aligned to a unified fusion time to complete the time synchronization of the data.

3. The method according to claim 2, characterized in that, Using the synchronized UAV operational status data as a known baseline, the relative motion state of the obstacle is estimated by fusing the data through a state estimation filtering algorithm. The collision time estimate and its standard deviation are then calculated based on the relative motion state. Specifically: Using the position, velocity, and attitude in the synchronized UAV operating state data as known reference inputs, and the relative position and velocity of the obstacle relative to the UAV as state variables to be estimated, the state equation of the state estimation filtering algorithm is established. The obstacle distance and azimuth angle observed by lidar, the obstacle depth map observed by visual camera, and the radial velocity observed by millimeter-wave radar are respectively established as corresponding observation equations; The observation data is sequentially updated according to the arrival order, and the relative position and relative velocity of the obstacle relative to the UAV are obtained by fusion estimation; The collision time estimate is calculated based on the ratio of the relative position to the relative velocity, and the estimated standard deviation of the collision time estimate is calculated based on the state covariance matrix output by the state estimation filtering algorithm.

4. The method according to claim 1, characterized in that, Based on the current flight speed, the maximum available deceleration, and the total delay from detection to braking, a first threshold for triggering emergency braking and a second threshold for triggering smooth braking are calculated, respectively, as follows: The first threshold is the sum of a first set ratio of the pure deceleration time and the total delay; the pure deceleration time is the ratio of the current flight speed to the maximum available deceleration. The second threshold is the sum of a second set ratio of the pure deceleration time and the total delay; and the second set ratio is greater than the first set ratio.

5. The method according to claim 1, characterized in that, The safe corrected collision time is determined by subtracting a first set multiple of the estimated standard deviation from the estimated collision time, and the resulting difference is taken as the safe corrected collision time.

6. The method according to claim 1, characterized in that, The process of generating the deceleration command is as follows: ； in, Indicates a deceleration command; Indicates the braking stiffness coefficient; Indicates the second threshold; This represents the estimated collision time. This represents the preset normalized time constant; This represents the smoothing factor.

7. The method according to claim 1, characterized in that, In the emergency braking mode, the optimal deceleration command is solved through constraint optimization with the reference trajectory as the target, specifically as follows: Using the reference trajectory as the deceleration tracking target, a multi-objective cost function is constructed that includes deceleration tracking error, attitude angle deviation, and thrust change rate. The constraints are limited by the speed of the UAV motor, the pitch angle, and the rate of change of thrust. Within each control cycle, with the goal of minimizing the cost function, the optimal control sequence in the preset time domain is solved within the feasible domain defined by the constraints, and the first control variable in the optimal control sequence is output as the optimal deceleration command for the current cycle.

8. The method according to claim 1, characterized in that, Methods for obtaining the current total mass of a drone include: During the takeoff and hovering phase, the initial mass is calculated by back-calculating the throttle and thrust calibration curves. During the level flight phase, the mass estimate is updated online based on the longitudinal thrust balance equation. During braking, the sliding average of the most recent valid mass estimate before braking is triggered is used as the current total mass of the UAV, and the mass change is compensated by integrating the battery current.

9. The method according to claim 1, characterized in that, The method further includes: when the relative velocity of an obstacle is detected to be less than or equal to zero and the relative distance to the obstacle continues to increase in multiple consecutive frames, it is determined that the collision risk has been eliminated, and the deceleration command exits the braking process according to an exponential decay law; the exponential decay law is specifically as follows: ； in, This indicates the deceleration command at the current moment; The deceleration command indicating exit from the start time. Indicates the current moment; Indicates the start time of exit; This represents the decay time constant.

10. An adaptive smooth obstacle avoidance and braking device for medium-to-large-sized unmanned aerial vehicles (UAVs), comprising at least one processor and a memory, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the at least one processor, it implements an adaptive smooth obstacle avoidance braking method for medium and large-sized unmanned aerial vehicles as described in any one of claims 1 to 9.

Images