Staying mechanism of airship robot and visual auxiliary staying and take-off control method

Abstract

The invention discloses a residence mechanism of an airship robot and a visual auxiliary residence and take-off control method. The airship robot man is composed of an air bag, a residence mechanism and a multi-rotor rack. The residing mechanism is composed of a balloon connecting base, a sliding block, a square nut, a threaded shaft micro motor, a motor-suction cup fixing base, an air valve, an air valve shell and a silica gel suction cup. An output shaft of the threaded shaft micro motor controls the square nut to move up and down so as to control opening and closing of the air valve. And the closing and opening of the air valve respectively correspond to the adsorption residence and separation take-off of the silica gel sucking disc. The visual auxiliary control system comprises a YOLOv8-based wall surface identification and positioning method and a residence and take-off control method. The residence control method comprises yaw angle control and distance control, and the take-off control method comprises pitch angle control and distance control. When the cruising ability of the airship robot is insufficient, the airship robot can reside on the wall surface for charging supply, and the cruising ability of the airship robot in scenes such as indoor environment detection and indoor emergency rescue is improved.

Description

Technical Field

[0001] This invention relates to the technical fields of robots, aircraft, airship robots, and dwell control, specifically to the dwelling mechanism of an airship robot and a vision-assisted dwelling and takeoff control method. Background Technology

[0002] In the current development of aerial work equipment, quadcopter drones, with their advantages of flexible control and environmental adaptability, have been widely used in logistics delivery, disaster relief, aerial photography, and other fields. However, their inherent limitations are becoming increasingly apparent: battery power supply limits endurance, making it difficult to support long-term operations; the high-speed rotating propellers generate significant noise and pose a safety threat to the surrounding environment and personnel, problems that are particularly prominent in enclosed environments such as indoor spaces. To overcome these bottlenecks, researchers have turned their attention to airships, a special flight platform. Airships rely on gases with a density lower than air to achieve levitation and flight, naturally possessing advantages such as high payload ratio and low energy consumption.

[0003] However, airships have relatively short flight times and are easily damaged when parked. A more attractive alternative is to allow them to remain stationary for a period of time between flights, like many small flying animals. Therefore, it is very useful to have airships alight on vertical surfaces such as building walls. Their passive attachment to these surfaces reduces the robot's own energy consumption, allowing them to remain as stable platforms for hours or even days. This makes them suitable for indoor surveillance, inspection, or indoor environmental rescue.

[0004] In recent years, perching technology has developed rapidly, often employing quadcopter drones equipped with perching devices to perform aerial perching tasks on various targets such as ceilings, tree branches, and walls. For example, Chinese invention patent application (CN201911158340.6) discloses a ceiling perching mechanism for a rotary-wing drone. The perching component includes an unlocking mechanism and a flexible adhesive attachment. The flexible adhesive attachment uses pressure-sensitive adhesive or a gecko-like dry adhesive material, combined with a silicone flexible joint for buffer alignment. Although this structure is simple, small in size, and produces no additional power consumption or noise in the perching state, the flexible adhesive attachment is a single-use component. After the drone returns, it remains on the ceiling, requiring reassembly for subsequent use, resulting in poor reusability. Chinese invention patent application (CN202111016724.1) provides a bistable perching mechanism for the top surface of a rotary-wing drone based on opposing claw spike units. Installed on the top surface of the drone, it consists of claw spike units, a bistable mechanism, a mounting base, a frame, servos, and a flexible connection mechanism. Although it employs a gripping design and can be reused multiple times without replacing consumables, its applicability is limited to surfaces like mineral wool ceilings where the grippers can pierce. It cannot be adapted to smooth walls, metal ceilings, or other non-porous, non-soft surfaces. Chinese invention patent application (CN202211391656.1) discloses a dual-arm roosting drone and its adaptive take-off, landing, and roosting method. During roosting, stability is achieved by adjusting the roosting arm deployment angle, the rotor stops for energy saving, and the roosting arm can be unlocked for detachment during go-around. While highly adaptable and covering various complex take-off and landing scenarios, it has strict limitations on the type and size of roosting points. It can only adapt to targets with specific structures such as lightning protection strips, cylinders, and slits, and cannot roost on smooth surfaces without attachment points, protrusions, or slits. Chinese invention patent application (CN 202310714022.3) discloses a biomimetic perching rotary-wing drone. The adsorption device consists of a detachable suction cup and a press-type spring latch, which are connected by a hinge mechanism and a press-type spring latch. Although the perching process requires no additional power consumption, the motion control is simple, and it can adapt to uneven surfaces, each perching requires controlling the drone's complex attitude transitions, making control difficult. Chinese invention patent application (CN202223558644.1) discloses an omnidirectional trigger-type drone perching device. When the elastic trigger deforms upon contact with the perching object, the air outlet connects to achieve vacuum adsorption; when detached, the trigger resets and blocks the opening, stopping the vacuum. Although it combines arc-shaped slider fine-tuning with trigger-type adsorption to meet the perching needs of both tilted and moving objects, it relies on a vacuum generator to provide adsorption power, resulting in high power consumption during perching.

[0005] In summary, existing loitering mechanisms either have complex loitering control strategies and are difficult to control during loitering, or the loitering mechanisms themselves are bulky, placing a heavy burden on the aircraft and making them inconvenient to carry in the air. There is an urgent need for a technical solution that can simplify loitering control strategies, reduce the burden on the aircraft during flight, and reduce energy consumption during loitering. This is a technical problem that needs to be solved by those in the field. Summary of the Invention

[0006] To address the aforementioned problems, this invention proposes a dwelling mechanism for an airship robot and a vision-assisted dwelling and takeoff control method.

[0007] To achieve the above objectives, the dwelling mechanism of the airship robot of the present invention includes an airbag, a multi-rotor frame, and a dwelling mechanism:

[0008] The multi-rotor frame is mounted on the bottom of the airbag, and the dwell mechanism is mounted on the equator of the airbag, facing the nose of the multi-rotor frame;

[0009] The multi-rotor frame includes a base plate, landing gear connectors, arm mounts, a flight control board, an arm, a motor mounting base, a blade protection ring, landing gear, lower base plate fixing studs, upper base plate fixing studs, an airbag connector, an onboard computer, a depth camera, a brushless motor, and a three-bladed rotor. The lower end of the upper base plate fixing stud is mounted on the base plate, and the upper end is fixedly connected to the airbag connector. The flight control board is fixed to the base plate with foam adhesive. The upper end of the lower base plate fixing stud is mounted on the base plate, and the lower end is fixedly connected to the onboard computer. The upper end of the landing gear connector is mounted below the base plate, and the lower end is fixedly connected to the landing gear. The arm mounts are mounted on the base plate, with one end of the arm fixedly connected to the arm mounts and the other end fixedly connected to the motor mounting base. The blade protection ring is mounted on the motor mounting base. The brushless motor is mounted in the middle of the blade protection ring. The three-bladed rotor is mounted on the center output shaft of the brushless motor. The depth camera is mounted on the base plate.

[0010] The dwelling mechanism includes a micro motor, a square nut, a slider, an air valve, an air valve housing, a silicone suction cup, a micro motor mounting base, a dwelling mechanism mounting base, and screws. The micro motor is mounted on the micro motor mounting base, the square nut is mounted on the lead screw output shaft of the micro motor, the slider is mounted on the square nut, the dwelling mechanism mounting base has four mounting holes for fixing four screws, the air valve is fixed in the air valve housing, the upper end of the air valve housing is fixed in the front circular hole of the micro motor mounting base, and the lower end is fixedly connected to the silicone suction cup. The screws fix the dwelling mechanism mounting base and the micro motor mounting base together.

[0011] As a further improvement to the dwelling mechanism of the airship robot of the present invention, the airbag is a spherical or other shaped helium or hydrogen balloon, and the multi-rotor frame is a hexacopter.

[0012] As a further improvement to the dwelling mechanism of the airship robot of this invention, the flight control board and the onboard computer are installed on the multi-rotor frame; the flight control board is responsible for collecting data from sensors such as gyroscopes and accelerometers, communicating with the onboard computer and ground station through the MAVLink protocol, transmitting flight status and control commands, and calculating the operation commands issued by the onboard computer into specific speed control for each motor through a mixer; the onboard computer is responsible for integrating depth camera data, completing image recognition, point cloud processing, and path planning tasks, and issuing control commands to the flight control board through the MAVLink protocol.

[0013] As a further improvement to the dwelling mechanism of the airship robot of the present invention, the square nut is installed on the output shaft of the micro motor through internal thread, the slider is a cuboid structure with a square hole on the right side for fixing the square nut and a cylindrical protrusion on the left side for pressing the compression spring of the air valve. The slider moves the output shaft of the micro motor to the central axis of the cylindrical protrusion. The dwelling mechanism has a rectangular groove below the fixed base, and the slider can move up and down in the groove.

[0014] See Figure 13 As shown, this invention provides a YOLOv8-based wall recognition, positioning, and dwelling control method for the dwelling mechanism of an airship robot, comprising the following steps:

[0015] (1) Data acquisition and preprocessing:

[0016] The depth camera outputs data collaboratively through dual sensors; the RGB sensor collects environmental texture information, and the depth sensor outputs pixel-level depth values; due to the difference in the field of view between the RGB and depth sensors, the depth map needs to be aligned to the RGB image coordinate system using the rs2::align tool; the alignment principle is based on the extrinsic parameter matrix calibrated at the factory of the depth camera, mapping the depth map pixels to the pixel grid of the RGB image through coordinate transformation, and finally outputting a depth map with the same size as the RGB image, ensuring that each RGB pixel corresponds one-to-one with the depth pixel; the intrinsic parameters of the depth camera are used to perform distortion correction on the RGB image and the depth map; to balance accuracy and computational efficiency, the image size is scaled to 640×480, and the intrinsic parameters are adjusted proportionally;

[0017] (2) Real-time segmentation of wall information:

[0018] The core of this module is to accurately segment the wall area from the preprocessed RGB image, providing a unique object of interest for subsequent localization. A lightweight YOLOv8-nano-seg model is used, which achieves real-time segmentation through a three-level network of "feature extraction-fusion-prediction": the backbone uses a CSPDarknet structure, extracting multi-scale features of the wall image through multi-layer convolution and residual connections; the neck uses a PAN-FPN structure, fusing high and low layer features to enhance the recognition of small targets and edge details; the head outputs three types of information: bounding box, class probability, and segmentation mask, a binary matrix where 1 represents wall pixels and 0 represents background; a confidence threshold of 0.5 is set during model inference to improve the frame rate and meet the real-time requirements of the airship robot's dynamic flight.

[0019] (3) 3D point cloud generation and optimization:

[0020] This module fuses the 2D wall segmentation results with depth information to generate a colored 3D point cloud, realizing the spatial mapping of the wall from the image plane to the camera coordinate system, providing a 3D coordinate basis for localization. The segmentation mask M is a binary image, where M(u,v)=1 represents a wall pixel, and (u,v) is the image coordinate. It is multiplied pixel by pixel with the aligned depth map D, retaining the depth value of the wall region D'(u,v)= D(u,v)∙M(u,v), and removing invalid depth values ​​from the background region. For pixels within the mask but with a depth value of 0, a 3×3 neighborhood mean is used for filling to avoid holes in the point cloud. For the processed depth map D', each valid pixel (u,v) (D'(u,v) ≠ 0) is traversed, and it is back-projected from the image coordinate system to the camera coordinate system based on the camera intrinsic parameters. The back-projection formula is as follows:

[0021]

[0022] Among them, (c x , c y (f) represents the principal point coordinates of the RGB camera. x , f y The focal length is provided by the camera's factory calibration; color values ​​(R, G, B) are extracted from the (u,v) positions of the RGB image and bound to the 3D coordinates (X, Y, Z) to generate a point cloud in (X, Y, Z, R, G, B) format; to reduce noise, statistical filtering is used to remove outliers, and the final output is the 3D point cloud P={p_{u,v} of the target wall. i | i=1,2, ∙∙∙, N}, where N is the number of point clouds;

[0023] (4) Distance calculation and proximity command generation:

[0024] The module uses point cloud computing to determine the relative position of the target wall and combines this with preset rules to generate an approach command for the airship robot, thus achieving a closed loop of perception-decision-control. The point cloud P of the target wall contains multiple 3D points, and its geometric center point can represent the equivalent position of the wall. The calculation method is the average of the coordinates of all points.

[0025]

[0026] Among them, (X) c ,Y c Z c Let (0, 0, 0) be the coordinates of the center point in the camera coordinate system; the position of the airship robot in the camera coordinate system is the origin (0, 0, 0), therefore the straight-line distance between the wall and the airship robot is: Meanwhile, through X c and Y c Determining the relative yaw angle φ and pitch angle θ simplifies to small angle approximations:

[0027]

[0028] The system generates quantitative control commands based on distance D and angle φ: when D > 1m, the airship robot moves along the Z-axis at normal operating speed while adjusting the φ yaw angle to ensure the center of the wall is in the center of the camera's field of view; when D < 1m, the forward speed is reduced to 0.1m / s, and the yaw angle is finely adjusted to reduce overshoot; at the same time, the micro motor controls the slider to move away from the compression spring of the air valve, completing the sealing task inside the air chamber of the silicone suction cup.

[0029] See Figure 6 As shown in Figures 7 and 8, the present invention provides a method for controlling the dwelling and takeoff of the dwelling mechanism of the airship robot, comprising the following steps:

[0030] (1) Stay control:

[0031] A dual-closed-loop PID structure is adopted, employing both distance control and attitude control. Distance information serves as the core objective of distance PID control, while the yaw angle, a key attitude parameter, is the adjustment target for attitude PID control. The distance detection module utilizes the aforementioned depth camera (2-13) to acquire the real-time actual distance d between the airship robot's fuselage and the wall surface. act The sampling frequency is set to 10Hz to ensure the timeliness of distance information; at the same time, the target distance d is preset. set This serves as the reference for the outer loop control; the distance PID control outer loop uses the deviation between the actual distance and the target distance as input, i.e., ∆d = d set -d act The attitude control command is output through the PID controller; the inner loop collects the airship's velocity v in the x-direction in real time. actSimultaneously preset the dwell speed v set This serves as the reference for the inner loop control. The speed PID control inner loop uses the deviation between the actual speed and the target speed as the input, i.e., ∆v = v set -v act The outer loop of the attitude PID control takes the deviation between the actual attitude and the target attitude as input, and collects the current actual yaw angle θ of the airship robot through the attitude sensor IMU. act Calculate the attitude deviation ∆f = f set - 𝜓 act The attitude PID control inner loop uses the yaw angle and actual angular velocity φ. zact angular velocity ω relative to the target yaw angle zset The deviation is taken as the input, and the actual yaw angle and angular velocity of the airship are collected by the attitude sensor IMU. zact Calculate the angular velocity deviation ∆r z =𝜔 zset -𝜔 zact The propellers of the airship robot adjust their rotation speed to control lift, thrust, and current attitude; after the actuators respond to the control signal, the airship robot's attitude changes, thus affecting its distance d from the wall. act Change; the distance detection module collects new d again. act The process of repeated deviation calculation, PID operation, and attitude adjustment forms a closed-loop control until ∆d stabilizes within the allowable error range of 50cm, thus achieving dwell stability.

[0032] (2) Takeoff control:

[0033] The entire process is based on attitude level changes, powered by rotor thrust, and fed back by distance information. The specific control logic is as follows: Takeoff control is triggered by a stable hovering state. When a takeoff command is received, the system first confirms the current tilt state of the airship robot using attitude sensors, and simultaneously records the actual hovering distance d using distance sensors. res And set the target takeoff distance d takeoff = 1m, serving as the reference for distance control during takeoff; the first stage involves adjusting the attitude from tilted to level; the task in this stage is to adjust the airship's pitch angle, using attitude PID control, but updating the target attitude parameters to the target pitch angle φ. set Set to 0°, target yaw angle 𝜓 set Maintain consistency with the status during the stay, i.e., ∆i = i set -𝜃 act ∆𝜓=0°; Attitude PID control inner loop uses pitch angle and actual angular velocity 𝜔 yact angular velocity ω relative to the target pitch angle ysetThe deviation is taken as the input, and the current actual pitch angle and angular velocity of the airship are collected by the attitude sensor IMU. yact Calculate the angular velocity deviation ∆r y =𝜔 yset -𝜔 yact According to ∆𝜔 y The output actuator signal causes the airship robot's propellers to adjust their rotation speed to control lift and current attitude, gradually counteracting the tilt angle; in the second stage, rotor thrust propels the airship robot away from the wall; when the attitude sensor detects the actual pitch angle θ... act Stable at |θ act When the horizontal attitude is ≤ 1°, the dwelling mechanism starts working, controlling the communication between the internal air chamber and the external atmospheric pressure, and the dwelling mechanism detaches from the wall; the rotor begins to propel the airship away from the wall; similar to the distance control method during dwell, the outer loop of the distance PID control is based on the actual distance d. act Distance d from the target takeoff The deviation is the input, i.e., ∆d takeoff = d takeoff -d act The attitude control command is calculated and output through the PID controller.

[0034] When the distance sensor detects d act Reaching d takeoff At this time, the system will automatically switch to cruise mode, while the inner loop attitude PID maintains θ. set = 0°、Ψ set The system is stable, completing the transition from stationary position to takeoff and finally to cruise. Compared with existing technologies, the advantages and beneficial effects of this invention are:

[0035] (1) The dwelling mechanism of the airship robot designed in this invention is mainly composed of a micro motor, an air valve and a silicone suction cup. The dwelling mechanism is small in weight, lightweight and easy to install, and will not put too much burden on the airship robot.

[0036] (2) The dwelling mechanism of the airship robot designed in this invention has a simple structure. During the dwelling process, the negative pressure environment inside the suction cup is passively formed by collision and squeezing. There is no need to use air pumps, vacuum generators and other continuously working devices. The power consumption is low during dwelling and take-off.

[0037] (3) The dwelling and takeoff control method designed in this invention has a simple control strategy. Compared with the UAV, which requires complex attitude transformation to achieve low-power stationarity on the wall, the dwelling and control method of the airship robot does not require complex attitude transformation algorithm, thereby improving the control response speed and control accuracy, and making the control process more convenient.

[0038] (4) The wall recognition, positioning and dwell control method based on YOLOv8 designed in this invention has high segmentation accuracy by using YOLOv8 multi-scale feature fusion and semantic segmentation capabilities to resist interference such as illumination and occlusion; combined with depth data to generate three-dimensional point cloud and wall mask information, the wall positioning error is small, providing accurate spatial reference for dwell. Attached Figure Description

[0039] Figure 1 This is a block diagram of the airship robot system equipped with a dwelling mechanism according to the present invention;

[0040] Figure 2 This is a block diagram of the multi-rotor frame of the airship robot of the present invention;

[0041] Figure 3 Stereoscopic view of the assembly of the dwelling mechanism of the present invention Figure 1 ;

[0042] Figure 4 Stereoscopic view of the assembly of the dwelling mechanism of the present invention Figure 2 ;

[0043] Figure 5 This is an exploded view of the dwelling mechanism of the present invention;

[0044] Figure 6 This is a block diagram of distance control during the dwell and takeoff control of the present invention;

[0045] Figure 7 This is a block diagram of the yaw angle control during the dwell control of the present invention;

[0046] Figure 8 This is a block diagram of the pitch angle control during takeoff control according to the present invention;

[0047] Figure 9 This is a schematic diagram of the dwelling process of the airship robot of the present invention;

[0048] Figure 10 This is a schematic diagram of an experimental scenario for the airship robot of the present invention during loitering and takeoff missions;

[0049] Figure 11 This is a mathematical model diagram of the dwelling system of the airship robot of the present invention during the dwelling process;

[0050] Figure 12 This is a laminar flow model of a flat plate in contact with a wall surface using a silicone suction cup, as described in this invention.

[0051] Figure 13 This is a flowchart of the wall recognition, positioning, and dwell control method based on YOLOv8 of the present invention.

[0052] The specific component names in the diagram are as follows:

[0053] 1. Airbag; 2. Multirotor frame; 3. Dwelling mechanism; 2-1. Base plate; 2-2. Landing gear connector; 2-3. Arm mounting bracket; 2-4. Flight control board; 2-5. Arm; 2-6. Motor mounting base; 2-7. Propeller blade protection ring; 2-8. Landing gear; 2-9. Lower base plate fixing stud; 2-10. Upper base plate fixing stud; 2-11. Airbag connector; 2-12. Onboard computer; 2-13. Depth camera; 2-14. Brushless motor; 2-15. Three-bladed propeller; 3-1. Micro motor; 3-2. Square nut; 3-3. Slider; 3-4. Air valve; 3-5. Air valve housing; 3-6. Silicone suction cup; 3-7. Micro motor mounting base; 3-8. Dwelling mechanism mounting base; 3-9. Screw. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific implementation of the dwelling process.

[0055] See Figure 1 As shown, the dwelling mechanism of the airship robot of the present invention includes an airbag 1, a multi-rotor frame 2, and a dwelling mechanism 3. The multi-rotor frame 2 is installed at the bottom of the airbag 1, and the dwelling mechanism 3 is installed on the equator of the airbag 1, facing the nose of the multi-rotor frame 2.

[0056] See Figure 2As shown, the multi-rotor frame 2 of the present invention includes a base plate 2-1, a landing gear connector 2-2, an arm fixing component 2-3, a flight control board 2-4, an arm 2-5, a motor mounting base 2-6, a propeller blade protection ring 2-7, a landing gear 2-8, a lower fixing stud 2-9 of the base plate, an upper fixing stud 2-10 of the base plate, an airbag connector 2-11, an onboard computer 2-12, a depth camera 2-13, a brushless motor 2-14, and a three-bladed propeller 2-15. The lower end of the upper fixing stud 2-10 is mounted on the base plate 2-1, and the upper end is fixedly connected to the airbag connector 2-11. The flight control board 2-4 is fixed to the base plate 2-1 with foam adhesive. The upper end of the lower fixing stud 2-9 is mounted on the base plate 2-1. The landing gear connector 2-2 is mounted on the base plate 2-1, with its lower end fixedly connected to the onboard computer 2-12. The upper end of the landing gear connector 2-2 is mounted below the base plate 2-1, and its lower end is fixedly connected to the landing gear 2-8. The arm fixing component 2-3 is mounted on the base plate 2-1. One end of the arm 2-5 is fixedly connected to the arm fixing component 2-3, and the other end is fixedly connected to the motor mounting base 2-6. The blade protection ring 2-7 is mounted on the motor mounting base 2-6. The brushless motor 2-14 is mounted in the middle of the blade protection ring 2-7. The three-bladed blade 2-15 is mounted on the center output shaft of the brushless motor 2-14. The depth camera 2-13 is mounted on the base plate 2-1.

[0057] See Figure 3 As shown in Figures 4 and 5, the dwelling mechanism 3 of the present invention includes a micro motor 3-1, a square nut 3-2, a slider 3-3, an air valve 3-4, an air valve housing 3-5, a silicone suction cup 3-6, a micro motor mounting base 3-7, a dwelling mechanism mounting base 3-8, and M3*30 screws 3-9. The micro motor 3-1 is mounted on the micro motor mounting base 3-7, the square nut 3-2 is mounted on the lead screw output shaft of the micro motor 3-1, the slider 3-3 is mounted on the square nut 3-2, the dwelling mechanism mounting base 3-8 has four mounting holes for fixing four M3*30 screws 3-9, the air valve 3-4 is fixed in the air valve housing 3-5, the upper end of the air valve housing 3-5 is fixed in the front end round hole of the micro motor mounting base 3-7, and the lower end is fixedly connected to the silicone suction cup 3-6. The M3*30 screws 3-9 fix the dwelling mechanism mounting base 3-8 and the micro motor mounting base 3-7 together.

[0058] During the initial contact of the silicone suction cup with the wall, the air valve is closed, and the gas inside the suction cup's air chamber can only escape through the edge of the suction cup. Due to the obstruction of the wall, the airship's speed gradually decreases until it reaches zero. Subsequently, due to the conservation of momentum, the airship will inevitably tend to move outwards from the wall. At this point, the edge of the suction cup is tightly adhered to the wall, and the internal air chamber of the suction cup is sealed, creating a negative pressure condition. Therefore, the collision process can be divided into the following two stages:

[0059] Phase 1: From the moment the suction cup edge first contacts the wall until the airship's overall speed reaches 0. During this phase, as the airship's direction of motion points inward towards the wall, the airship's balloon and the silicone suction cup of the dwell mechanism are compressed, and the gas inside the suction cup is expelled from its edge.

[0060] Phase Two: From the moment the airship's overall speed reaches 0 until it reaches a steady state where it is stationary relative to the wall. During this phase, the edges of the suction cups are tightly pressed against the wall, the air chambers inside the suction cups are sealed, and the airship's balloons and the silicone suction cups of the dwell mechanism repeatedly experience compression and stretching, exhibiting an oscillating trend.

[0061] (1) Mathematical modeling of the first stage of the resident system;

[0062] Figure 9 The diagram illustrates the airship's hovering process, describing the physical systems involved in airship hovering. Figure 10 The diagram illustrates an experimental scenario of an airship robot performing hovering and takeoff missions. Modeling such a real physical system is difficult, requiring numerous model parameters and resulting in a heavy workload. Therefore, the mathematical model of the hovering process is reasonably simplified. The total mass of the airship and hovering mechanism is denoted by m; the airbag is approximated as having an elastic coefficient of k. s The spring; the silicone suction cup can be approximated as a cylindrical piston F. suction And the elastic coefficient is k p The combination of springs; in stage one, k can be approximated. s and k p It is considered as a series connection. The simplified model of the resident system is as follows: Figure 11 As shown.

[0063] Analyzing block m, we can obtain the following from Newton's second law:

[0064]

[0065] Where k is k s and k p The equivalent spring constant of the series-connected springs is given by k, ΔP is the change in air pressure inside the suction cup, A is the area of ​​the bottom surface of the suction cup, and x is the deformation of the series-connected equivalent springs. The expressions for k, x, and ΔP are shown below:

[0066]

[0067]

[0068]

[0069] Where, x s and x p spring k s and k p The deformation is given by P(t), where P(t) is the pressure inside the suction cup air cavity at time t, and P0 is the standard atmospheric pressure.

[0070] Analysis of the gas inside the suction cup's gas chamber: The gas inside the suction cup's gas chamber is air, which can be approximated as an ideal gas at room temperature and pressure. According to the ideal gas law, we can obtain:

[0071]

[0072] Where V(t) is the volume of the air cavity inside the suction cup at time t, and R air Let be the ideal gas constant, T be the thermodynamic temperature of the ideal gas, and Q be the exhaust mass flow rate. The formula means that the rate of change of gas mass within the suction cup is equal to the exhaust mass flow rate. Expanding under isothermal conditions (T constant):

[0073]

[0074] The air cavity inside the suction cup is shaped like a frustum. To simplify the model and facilitate analysis, the frustum is simplified to a cylinder with equal volume and height. Based on the formulas for the volume of a frustum and a cylinder, the relationship between the base area A' of the equivalent cylinder and the radius r of the top surface and the radius R of the bottom surface of the frustum can be derived:

[0075]

[0076] Among them, h p Let A' be the height of the air cavity inside the suction cup, and R be the base area of ​​the equivalent cylinder. Then, r and R are the radii of the top and bottom surfaces of the frustum, respectively. Therefore, V(t) can be calculated using the following formula:

[0077]

[0078] When the suction cup first contacts the wall, the air outlet of the suction cup is sealed by the air valve, and the gas inside the suction cup's air chamber can only flow out through the gap between the suction cup and the wall. Figure 12As shown. Therefore, the suction cup and the wall can be considered as two parallel plates, with a gap of H. The wall is located at y=0, the edge of the suction cup is located at y=H, and the width of the plate in the direction of airflow is W. Air is considered an incompressible fluid, and the airflow is a steady-state laminar flow, meaning the air velocity u only varies along the y-direction. There is no boundary slip condition, meaning the surface velocity of the plate is zero. Along... The pressure gradient along the x-axis is considered constant. Based on the above physical model and assumptions, the Navier-Stokes equations along the x-axis can be derived:

[0079]

[0080] Where ρ is the air density, u, v, and w are the velocity components of the fluid in the x, y, and z directions, respectively, ∂P / ∂x is the pressure gradient in the x-direction, and μ is the dynamic viscosity of the air. Based on the steady-state laminar flow assumption ∂u / ∂t=0, and the velocity only varies along the y-direction (i.e., ∂u / ∂x=∂u / ∂z=0), and the vertical velocity v=w=0, the equation can be simplified to:

[0081]

[0082] Based on the assumption of a constant pressure gradient, the pressure gradient can be obtained as follows:

[0083]

[0084] Where L is the length of the pressure drop in the x-direction, we can obtain:

[0085]

[0086] Solving this differential equation yields:

[0087]

[0088] Where C1 and C2 are integration constants. Combining the no-slip boundary conditions:

[0089]

[0090] C1 and C2 can be obtained:

[0091]

[0092] Therefore, the velocity distribution is as follows:

[0093]

[0094] It can be seen that the velocity distribution along the y-direction follows a parabolic equation. The velocity is zero at the surface of the plate, and reaches its maximum value at the midpoint of the gap between the plates, i.e., at H / 2. The flow rate Q is the integral of the velocity over the width W of the parallel plates.

[0095]

[0096] The Poiseuille equations for a laminar flow model of a flat plate in contact with a suction cup and a wall can be obtained:

[0097]

[0098] In the laminar flow model of a flat plate in contact with a suction cup and a wall, the length L of the pressure drop in the x-direction is the radius R of the bottom surface of the suction cup, and the width W of the parallel plate is the side surface area of ​​the cylindrical gap between the suction cup and the wall.

[0099]

[0100] Available traffic Q:

[0101] (2-26)

[0102] By simultaneously solving (2-7), (2-8), (2-10), (2-12), (2-14), and (2-26), and rearranging, we can obtain the dynamic model of stage one of the residence process:

[0103]

[0104] This model allows us to derive the system state at the moment when the velocities of the airship and suction cup are initially zero: spring k p The shape variable x p0 Spring k s The shape variable x s0 And the air pressure P0' inside the suction cup air chamber. These system states will serve as the initial states for stage two of the residence process.

[0105] (2) Mathematical modeling of the second stage of the resident system;

[0106] Phase two of the dwell process spans from the moment the airship's overall velocity reaches zero until it reaches a steady state where it is stationary relative to the wall. Before mathematically modeling phase two, it's necessary to analyze whether the suction cups and the wall can maintain contact. If the suction cups cannot maintain adhesion as the airship tends to move outwards from the wall, then the mathematical model for phase two will be meaningless. Therefore, determining whether the suction cups and the wall can maintain contact is a necessary prerequisite for mathematical modeling phase two.

[0107] For example, in a dwell system model, to analyze the contact relationship between the suction cup and the wall, points can be used... The system on the right simplifies to a force F of unknown magnitude acting at point U. At point U, according to Newton's second law, we have:

[0108]

[0109] Where ∆P' is the pressure difference between the external atmospheric pressure of the suction cup and the internal pressure of the suction cup's air cavity:

[0110]

[0111] The gas inside the suction cup's gas chamber is air, which can be approximated as an ideal gas at room temperature and pressure. The temperature inside the gas chamber can be approximated as constant. According to the ideal gas law, we can obtain:

[0112]

[0113] Where V(t) and V0' are respectively:

[0114]

[0115]

[0116] And by organizing, we can obtain:

[0117]

[0118] This is about x p The equation, x p Let x be the deformation of the suction cup, and its value range is x. p >−h p .make

[0119]

[0120] Observe the function f(x) p We can obtain:

[0121]

[0122]

[0123]

[0124] By the Intermediate Point Theorem, we can obtain that the function f(x) p ) in x p >−h p There must exist a zero point within the range. Therefore, there must exist a reasonable real root that satisfies the equation. Thus, the suction cup and the wall can always maintain contact.

[0125] Based on the above analysis, the suction cup can remain adhered to the wall, which prepares the ground for the analysis of stage two of the dwelling process. In stage two of the dwelling process, applying Newton's second law to block m yields:

[0126]

[0127] Applying Newton's second law to point U, we get:

[0128]

[0129] Among them, c eq Let be the equivalent viscous damping coefficient of the suction cup. Since the damping of the suction cup is complex and difficult to calculate accurately directly, the equivalent viscous damping method is used to represent all damping effects as a single coefficient. The dynamic characteristics of the suction cup are then analyzed using classical vibration theory.

[0130]

[0131] Where Ψ is the empirical damping coefficient, f is the dynamic frequency of the chuck operation, and k dyn Dynamic stiffness describes the elastic force generated per unit deformation of the suction cup under dynamic load:

[0132]

[0133] Among them, E dyn This is the dynamic elastic modulus of the suction cup, where A is the area of ​​the lower surface of the suction cup, and h is the dynamic elastic modulus of the suction cup. p Let be the initial height of the air cavity inside the suction cup. Combining and rearranging these equations, we obtain the dynamic model for stage two of the residence process:

[0134]

[0135] In this way, dynamic models of stage one and stage two of the residence process were established, providing a theoretical basis for residence control.

[0136] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A dwelling mechanism for an airship robot, comprising an airbag (1), a multi-rotor frame (2), and a dwelling mechanism (3), characterized in that: The multi-rotor frame (2) is installed at the bottom of the airbag (1), and the dwelling mechanism (3) is installed on the equator of the airbag (1) and faces the nose of the multi-rotor frame (2); The multi-rotor frame (2) includes a base plate (2-1), landing gear connector (2-2), arm mounting bracket (2-3), flight control board (2-4), arm (2-5), motor mounting base (2-6), blade protection ring (2-7), landing gear (2-8), bottom mounting stud (2-9), top mounting stud (2-10), airbag connector (2-11), onboard computer (2-12), depth camera (2-13), brushless motor (2-14), and three-bladed rotor (2-15). The bottom end of the top mounting stud (2-10) is mounted on the base plate (2-1), and the top end is fixedly connected to the airbag connector (2-11). The flight control board (2-4) is fixed to the base plate (2-1) with foam adhesive. The top end of the bottom mounting stud (2-9) is fixed to the base plate (2-1). The landing gear connector (2-2) is mounted on the base plate (2-1), with its lower end fixedly connected to the onboard computer (2-12). The upper end of the landing gear connector (2-2) is mounted below the base plate (2-1), and its lower end is fixedly connected to the landing gear (2-8). The boom fixing component (2-3) is mounted on the base plate (2-1). One end of the boom (2-5) is fixedly connected to the boom fixing component (2-3), and the other end is fixedly connected to the motor fixing base (2-6). The blade protection ring (2-7) is mounted on the motor fixing base (2-6). The brushless motor (2-14) is mounted in the middle of the blade protection ring (2-7). The three-bladed blade (2-15) is mounted on the center output shaft of the brushless motor (2-14). The depth camera (2-13) is mounted on the base plate (2-1). The dwelling mechanism (3) includes a micro motor (3-1), a square nut (3-2), a slider (3-3), an air valve (3-4), an air valve housing (3-5), a silicone suction cup (3-6), a micro motor mounting base (3-7), a dwelling mechanism mounting base (3-8), and a screw (3-9); the micro motor (3-1) is mounted on the micro motor mounting base (3-7), the square nut (3-2) is mounted on the lead screw output shaft of the micro motor (3-1), and the slider (3-3) is mounted on the... The square nut (3-2) has four mounting holes on the dwelling mechanism fixing base (3-8) for fixing four screws (3-9). The air valve (3-4) is fixed in the air valve housing (3-5). The upper end of the air valve housing (3-5) is fixed in the front round hole of the micro motor fixing base (3-7), and the lower end is fixedly connected to the silicone suction cup (3-6). The screws (3-9) fix the dwelling mechanism fixing base (3-8) and the micro motor fixing base (3-7) together.

2. The dwelling mechanism of the airship robot according to claim 1, characterized in that: The airbag (1) is a spherical or other shaped helium or hydrogen balloon, and the multi-rotor frame (2) is a hexacopter.

3. The dwelling mechanism of the airship robot according to claim 1, characterized in that: The multi-rotor frame (2) is equipped with the flight control board (2-4) and the onboard computer (2-12). The flight control board (2-4) is responsible for collecting data from sensors such as gyroscopes and accelerometers, communicating with the onboard computer (2-12) and the ground station through the MAVLink protocol, transmitting flight status and control commands, and calculating the operation commands issued by the onboard computer into specific speed control for each motor through a mixer. The onboard computer (2-12) is responsible for integrating depth camera data, completing image recognition, point cloud processing, and path planning tasks, and issuing control commands to the flight control board (2-4) through the MAVLink protocol.

4. The dwelling mechanism of the airship robot according to claim 1, characterized in that: The square nut (3-2) is installed on the output shaft of the micro motor (3-1) by internal thread. The slider (3-3) is a cuboid structure with a square hole on the right side for fixing the square nut (3-2) and a cylindrical protrusion on the left side for pressing the compression spring of the air valve (3-4). The slider (3-3) moves the output shaft of the micro motor (3-1) to the central axis of the cylindrical protrusion. The base (3-8) of the dwelling mechanism has a rectangular groove below it, and the slider (3-3) can move up and down in the groove.

5. The YOLOv8-based wall recognition, positioning, and dwelling control method for the dwelling mechanism of the airship robot according to any one of claims 1-4, characterized in that, Includes the following steps: (1) Data acquisition and preprocessing: The depth camera (2-13) outputs data through the collaborative output of dual sensors; the RGB sensor collects environmental texture information, and the depth sensor outputs pixel-level depth values; since there is a difference between the field of view of the RGB and depth sensors, the depth map needs to be aligned to the RGB image coordinate system using the rs2::align tool. The alignment principle is based on the extrinsic parameter matrix calibrated at the factory of the depth camera (2-13). The depth map pixels are mapped to the pixel grid of the RGB image through coordinate transformation, and finally outputs a depth map with the same size as the RGB image, ensuring that each RGB pixel corresponds one-to-one with the depth pixel. The intrinsic parameters of the depth camera (2-13) are used to perform distortion correction on the RGB image and the depth map. To balance accuracy and computational efficiency, the image size is scaled to 640×480, and the intrinsic parameters are adjusted proportionally. (2) Real-time segmentation of wall information: The core of this module is to accurately segment the wall area from the preprocessed RGB image, providing a unique object of interest for subsequent localization; the YOLOv8-nano-seg lightweight model is selected, which achieves real-time segmentation through a three-level network of "feature extraction-fusion-prediction"; the backbone adopts the CSPDarknet structure, which extracts multi-scale features of the wall image through multi-layer convolution and residual connections. The Neck adopts a PAN-FPN structure, which integrates high and low layer features to enhance the recognition ability of small targets and edge details; the Head outputs three types of information: bounding box, class probability, and segmentation mask; the confidence threshold is set to 0.5 during model inference to improve the frame rate and meet the real-time requirements of the airship robot's dynamic flight. (3) 3D point cloud generation and optimization: This module fuses the 2D wall segmentation results with depth information to generate a colored 3D point cloud, realizing the spatial mapping of the wall from the image plane to the camera coordinate system, providing a 3D coordinate basis for positioning; The segmentation mask M is a binary image, where M(u,v)=1 represents a wall pixel, and (u,v) is the image coordinate. It is multiplied pixel-by-pixel by the aligned depth map D, retaining the depth value of the wall region D'(u,v) = D(u,v)∙M(u,v), while discarding invalid depth values ​​from the background region. For pixels within the mask but with a depth value of 0, a 3×3 neighborhood mean is used for filling to avoid point cloud holes. For the processed depth map D', each valid pixel (u,v) (D'(u,v) ≠ 0) is traversed, and it is back-projected from the image coordinates to the camera coordinate system based on camera intrinsics. The back-projection formula is as follows: ； Among them, (c x , c y (f) represents the principal point coordinates of the RGB camera. x , f y The focal length is provided by the camera's factory calibration; color values ​​(R, G, B) are extracted from the (u,v) positions of the RGB image and bound to the 3D coordinates (X, Y, Z) to generate a point cloud in (X, Y, Z, R, G, B) format; to reduce noise, statistical filtering is used to remove outliers, and the final output is the 3D point cloud P={p_{u,v} of the target wall. i | i=1,2, ∙∙∙, N}, where N is the number of point clouds; (4) Distance calculation and proximity command generation: The module uses point cloud computing to determine the relative position of the target wall and combines this with preset rules to generate an approach command for the airship robot, thus achieving a closed loop of perception-decision-control. The point cloud P of the target wall contains multiple 3D points, and its geometric center point can represent the equivalent position of the wall. The calculation method is the average of the coordinates of all points. ； Among them, (X) c ,Y c Z c Let (0, 0, 0) be the coordinates of the center point in the camera coordinate system; the position of the airship robot in the camera coordinate system is the origin (0, 0, 0), therefore the straight-line distance between the wall and the airship robot is: Meanwhile, through X c and Y c Determining the relative yaw angle φ and pitch angle θ simplifies to small angle approximations: ； The system generates quantitative control commands based on distance D and angle φ: when D > 1m, the airship robot moves forward along the Z-axis at normal operating speed, while adjusting the φ yaw angle to ensure that the center of the wall is located in the center of the camera's field of view; when D < 1m, the forward speed is reduced to 0.1m / s, and the yaw angle is finely adjusted to reduce overshoot; at the same time, the micro motor (3-1) controls the slider (3-3) to move away from the compression spring of the air valve (3-4) to complete the sealing task inside the air chamber of the silicone suction cup (3-6).

6. The method for controlling the dwelling and takeoff of the dwelling mechanism of the airship robot according to any one of claims 1-4, characterized in that, Includes the following steps: (1) Stay control: A dual-closed-loop PID structure is adopted, employing both distance control and attitude control. Distance information serves as the core objective of distance PID control, while the yaw angle, a key attitude parameter, is the adjustment target for attitude PID control. The distance detection module utilizes the aforementioned depth camera (2-13) to acquire the real-time actual distance d between the airship robot's fuselage and the wall. act The sampling frequency is set to 10Hz to ensure the timeliness of distance information; at the same time, the target distance d is preset. set This serves as the reference for the outer loop control; the distance PID control outer loop uses the deviation between the actual distance and the target distance as input, i.e., ∆d = d set -d act The attitude control command is output through the PID controller; the inner loop collects the airship's velocity v in the x-direction in real time. act Simultaneously preset the dwell speed v set This serves as the reference for the inner loop control. The speed PID control inner loop uses the deviation between the actual speed and the target speed as the input, i.e., ∆v = v set -v act The outer loop of the attitude PID control takes the deviation between the actual attitude and the target attitude as input, and collects the current actual yaw angle θ of the airship robot through the attitude sensor IMU. act Calculate the attitude deviation ∆f = f set - 𝜓 act The attitude PID control inner loop uses the yaw angle and actual angular velocity φ. zact angular velocity ω relative to the target yaw angle zset The deviation is taken as the input, and the actual yaw angle and angular velocity of the airship are collected by the attitude sensor IMU. zact Calculate the angular velocity deviation ∆r z =𝜔 zset -𝜔 zact The propellers of the airship robot adjust their rotation speed to control lift, thrust, and current attitude; after the actuators respond to the control signal, the airship robot's attitude changes, thus affecting its distance d from the wall. act Change; the distance detection module collects new d again. act The process of repeated deviation calculation, PID operation, and attitude adjustment forms a closed-loop control until ∆d stabilizes within the allowable error range of 50cm, thus achieving dwell stability. (2) Takeoff control: The entire process is based on attitude level changes, powered by rotor thrust, and fed back by distance information. The specific control logic is as follows: Takeoff control is triggered by a stable hovering state. When a takeoff command is received, the system first confirms the current tilt state of the airship robot using attitude sensors, and simultaneously records the actual hovering distance d using distance sensors. res And set the target takeoff distance d takeoff = 1m, serving as the reference for distance control during takeoff; the first stage involves adjusting the attitude from tilted to level; the task in this stage is to adjust the airship's pitch angle, using attitude PID control, but updating the target attitude parameters to the target pitch angle φ. set Set to 0°, target yaw angle 𝜓 set Maintain consistency with the status during the stay, i.e., ∆i = i set - 𝜃 act ∆𝜓=0°; Attitude PID control inner loop uses pitch angle and actual angular velocity 𝜔 yact angular velocity ω relative to the target pitch angle yset The deviation is taken as the input, and the current actual pitch angle and angular velocity of the airship are collected by the attitude sensor IMU. yact Calculate the angular velocity deviation ∆r y =𝜔 yset -𝜔 yact According to ∆𝜔 y The output actuator signal causes the airship robot's propellers to adjust their rotation speed to control lift and current attitude, gradually counteracting the tilt angle; in the second stage, rotor thrust propels the airship robot away from the wall; when the attitude sensor detects the actual pitch angle θ... act Stable at |θ act When the horizontal attitude is ≤ 1°, the dwelling mechanism starts working, controlling the communication between the internal air chamber and the external atmospheric pressure, and the dwelling mechanism detaches from the wall; the rotor begins to propel the airship away from the wall; similar to the distance control method during dwell, the outer loop of the distance PID control is based on the actual distance d. act Distance d from the target takeoff The deviation is the input, i.e., ∆d takeoff = d takeoff -d act The attitude control command is calculated and output through the PID controller. When the distance sensor detects d act Reaching d takeoff At this time, the system will automatically switch to cruise mode, while the inner loop attitude PID maintains θ. set = 0°、Ψ set Stable, completing the transition from stationary position to takeoff and finally to cruise.

Images