An amphibious robot and a control method thereof

Abstract

The application relates to the technical field of robots, and provides an amphibious robot and a control method thereof, the amphibious robot comprising a bottom plate, a first driving motor, a flange plate, a guide arm, a steering engine and a paddle wheel; the middle position of the front part of the bottom plate is provided with a rectangular opening, and the first driving motor is arranged on one side of the rectangular opening; the output shaft of the first driving motor is connected with the input end of the flange plate; the flange plate is located above the rectangular opening of the bottom plate; the output end of the flange plate is connected with the guide arm; the guide arm is T-shaped; the rear ends of the bottom plate and the two ends of the guide arm are respectively provided with the steering engine; the second motor is arranged on the steering engine; and the output end of the second motor is provided with the paddle wheel. The application can improve the motion stability, the anti-overturning capability, the slip control accuracy, the path planning safety and the cross-medium positioning reliability of the robot.

Description

Technical Field

[0001] This invention relates to the field of robotics, and in particular to an amphibious robot and its control method. Background Technology

[0002] Amphibious robots are intelligent equipment capable of performing tasks autonomously or semi-autonomously in both aquatic and terrestrial environments. They are characterized by strong environmental adaptability and diverse functions, and are widely used in fields such as polar scientific research, under-ice environment monitoring, and underwater search and rescue.

[0003] Most existing amphibious robots adopt snake-like, legged, or wheel-propeller switching structures. Among them, snake-like robots achieve movement through multi-joint swinging, which can adapt to complex terrain, but their movement speed is slow, control is complex, and energy efficiency is low; legged robots cross obstacles by walking on multiple legs, but their structure has high redundancy, is difficult to control, and has poor underwater stability; existing ordinary wheel-propeller switching robots can only switch propulsion modes through servo motors, and lack a pitch control mechanism at the front, so they cannot actively guide the body attitude when transitioning between ice and water.

[0004] In environmental monitoring in polar regions, frozen lakes, and marine areas, traditional amphibious robots struggle to perform continuous operations on ice and underwater, generally exhibiting the following shortcomings:

[0005] 1) The propulsion mode switching mechanism is single-function and lacks a coordinated mechanism for obstacle crossing and center of gravity adjustment; the front lacks an active pitch control mechanism, making the transition at the ice-water interface difficult.

[0006] 2) The entire machine lacks contact force sensing, making it impossible to perceive the wheel-ground contact status in real time; the ice thickness detection mechanism is missing, making it impossible to identify thin ice areas; the configuration of cross-media sensing sensors is insufficient, resulting in a low level of intelligence.

[0007] 3) It is prone to overturning at the ice-water interface transition and lacks active anti-overturning control; the path planning does not consider the ice layer bearing capacity and cannot avoid thin ice areas.

[0008] Therefore, there is an urgent need for an amphibious robot with structural coordination, active center of gravity adjustment, full-machine force perception, ice thickness detection, cross-medium autonomous identification, and ice-adaptive path planning to achieve safe, continuous, and efficient operation in ice and underwater environments. Summary of the Invention

[0009] This invention primarily addresses the technical problems of existing amphibious robots, such as the lack of a pitch control mechanism at the front, poor structural coordination, lagging overturning control, simplistic sliding control, insufficient path planning safety, and discontinuous cross-medium positioning. It proposes an amphibious robot and its control method to improve the robot's motion stability, anti-overturning capability, sliding control accuracy, path planning safety, and cross-medium positioning reliability.

[0010] This invention provides an amphibious robot, comprising: a base plate, a first drive motor, a flange, a guide arm, a servo motor, and a propeller wheel;

[0011] The base plate has a rectangular opening at the center of its front portion, and a first drive motor is mounted on one side of the rectangular opening. The output shaft of the first drive motor is connected to the input end of the flange. The flange is located above the rectangular opening of the base plate. The output end of the flange is connected to a guide arm. The guide arm is T-shaped.

[0012] Servo motors are installed at both ends of the rear portion of the base plate and at both ends of the guide arm; a second motor is installed on the servo motor, and a propeller wheel is installed at the output end of the second motor; the servo motor can drive the propeller wheel to rotate 90° to achieve switching between wheel mode and propeller mode.

[0013] Preferably, the propeller wheel has a plurality of blades evenly arranged, and the edge of the propeller wheel is embedded with blades.

[0014] Preferably, a housing is fitted above the base plate;

[0015] The housing has an elongated hole at a position corresponding to the guide arm;

[0016] The base plate is bonded to the outer surface of the casing with buoyancy foam.

[0017] The lower surface of the base plate is reinforced with reinforcing profiles in multiple locations.

[0018] Preferably, a square box is provided at the rear center of the base plate, and the battery and main controller are installed inside the square box;

[0019] A small box is installed in the middle of the lower surface of the base plate, and a gyroscope and a water pressure sensor are installed inside the small box.

[0020] Preferably, the front of the housing is equipped with a binocular vision camera, a millimeter-wave radar, and an acoustic ice thickness sensor;

[0021] Conductivity sensors are respectively installed on both sides of the base plate;

[0022] A temperature and humidity sensor is installed on the casing;

[0023] A wheel axle contact force sensor is installed on the propeller wheel.

[0024] Correspondingly, the present invention also provides a control method for an ice-water amphibious robot according to any embodiment of the present invention, comprising the following processes:

[0025] Step 1: Collect robot's operational and environmental data in real time. The operational data includes pitch angle, roll angle, force distribution on the four axles, wheel speed, three-axis acceleration, three-axis angular velocity, and body depth. The environmental data includes forward depth map, ice thickness, internal ice structure, conductivity, ambient temperature, and ambient humidity.

[0026] Step 2: Based on the real-time collected robot operation data and environmental data, determine the robot's posture stability, overturning risk level, slip type, accessibility level, and whether it crosses media, and then perform robot posture control based on the posture stability, overturning risk level, slip type, accessibility level, and whether it crosses media.

[0027] Step 2 includes the following steps 201 to 205:

[0028] Step 201: Determine the robot's attitude stability state using the instantaneous center of gravity position model, and perform attitude control on the robot based on the attitude stability state.

[0029] Step 202: Determine the overturning risk level of the robot using the LSTM overturning prediction model, and control the robot's posture based on the overturning risk level.

[0030] Step 203: Determine the robot's slip type using the ice surface slip classification model, and control the robot's posture based on the slip type;

[0031] Step 204: Determine the robot's navigability level using the ice layer bearing capacity model, and control the robot's posture based on the navigability level.

[0032] Step 205: Determine whether the robot is crossing media using the cross-media switching model, and control the robot's posture based on whether it is crossing media.

[0033] Step 3: Based on the real-time collected environmental data of the robot, perform multi-objective path planning using a multi-objective path planning model.

[0034] Furthermore, in step 201, the contact force values ​​of the four wheel axles and the coordinates of the guide arm's center of mass are... , , ), coordinates of the centroid of the square box ( , , Input the instantaneous center of gravity position model to obtain the instantaneous center of gravity coordinates. , ), center of gravity offset (ΔX, ΔY), stability margin value Based on the center of gravity offset, stability margin value, and preset stability boundary value, the guide arm pitch angle is adjusted, the differential torque distribution of the four second motors is adjusted, and the body attitude is corrected in a closed loop.

[0035] In step 202, the LSTM overturning prediction model adopts a four-layer network structure, including: the first layer is an LSTM recurrent neural network layer; the second layer is a Dropout layer; the third layer is a fully connected layer using the ReLU activation function; and the fourth layer is an output layer using the Sigmoid activation function.

[0036] The input to the LSTM overturning prediction model is the past Pitch and roll angles sampled within a time period Four axle contact force values ​​F1~F4, four wheel speeds ω1~ω4; the data collection time interval is... ;

[0037] The output of the LSTM overturning prediction model: Future Probability of overturning within a time period Overturning direction Estimated time to capsize ;

[0038] The attitude control of the LSTM overturning prediction model: when the overturning probability When the overturning probability threshold 'a' is reached, the guide arm swings in the opposite direction of overturning. At the same time, the torque k of the second motor on the overturning side is reduced.

[0039] Furthermore, step 203 includes steps 2031 to 2033:

[0040] Step 2031, the ice surface slip classification model calculates each The slip detection parameters for the time interval include wheel speed change rate Δω, fuselage displacement change rate Δv, slip rate s, roll angle change rate Δφ, electrical conductivity EC, and left and right wheel speed difference.

[0041] The formula for calculating the wheel speed change rate Δω:

[0042] Δω = ;

[0043] in, This represents the wheel speed at the current moment. The wheel speed is the speed at the previous moment; Δt is the sampling time interval.

[0044] Formula for calculating the rate of change of fuselage displacement Δv:

[0045] Δv = ∫ a(t) dt;

[0046] Where a(t) is the fuselage acceleration measured by the IMU; dt is the integration time step;

[0047] The formula for calculating slip ratio s:

[0048] s = ;

[0049] Wherein, the slip ratio s is dimensionless and ranges from 0 to 1;

[0050] The formula for calculating the roll angle change rate Δφ is as follows:

[0051] Δφ = ;

[0052] in, The current roll angle; The roll angle at the previous moment;

[0053] The conductivity EC is collected in real time by four conductivity sensors on both sides of the base plate;

[0054] Formula for calculating the speed difference between the left and right wheels:

[0055] | - | = | |;

[0056] in, Indicates the rotational speed of the two wheels on the left; Indicates the rotational speed of the two wheels on the right side; , , , These are the wheel speeds of the left front rotor, the right front rotor, the left rear rotor, and the right rear rotor, respectively.

[0057] Step 2032: Determine the type of ice surface slippage based on the calculated slippage detection parameters;

[0058] When the slip ratio s > the first slip ratio threshold and the fuselage displacement change rate Δv < the first fuselage displacement change rate threshold, the slip type is determined to be acceleration slip; when the fuselage displacement change rate Δv > the second fuselage displacement change rate threshold and the left and right wheel speed difference | - When the speed difference between the left and right wheels reaches the threshold, the slip type is determined to be lateral slip; when the conductivity EC changes abruptly within the range of 100~1000 μS / cm and the slip rate s > the second slip rate threshold, the slip type is determined to be ice-water interface slip; when none of the above conditions are met, the slip type is determined to be normal driving.

[0059] Step 2033: If the slip type is determined to be acceleration slip, pulse propulsion is used; if the slip type is determined to be lateral slip, differential reversal and guide arm side swing are used; if the slip type is determined to be ice-water interface slip, advance switching of propulsion mode and guide arm downward swing are used; if the slip type is determined to be normal driving, conventional PI control is used.

[0060] Furthermore, step 204 includes the following steps 2041 to 2044:

[0061] Step 2041: The ice layer bearing capacity model determines the thickness h of the ice layer in front based on the echo signal detected by the millimeter-wave radar and the pulse reflection time of the ice layer acoustic thickness sensor.

[0062] Step 2042: The ice layer bearing capacity model calculates the ice layer bearing capacity based on the real-time temperature data T and humidity data RH collected by the temperature and humidity sensor, combined with the ice layer thickness h.

[0063] The formula for calculating the bearing capacity of the ice layer is:

[0064] = k · h · (1 + 0.02 · (T - (-10)));

[0065] in, is the ice layer bearing capacity; k is the calibration coefficient; h is the ice layer thickness; T is the ambient temperature; -10 is the reference temperature;

[0066] Step 2043, the ice layer bearing capacity model is based on the calculated ice layer bearing capacity. The passability classification is determined based on the ice thickness h; the passability classification includes three levels: safe zone, warning zone, and restricted zone.

[0067] Step 2044: The ice layer bearing capacity model performs robot posture control based on the determined accessibility classification;

[0068] When the accessibility classification determines that the area is safe, the robot maintains its normal speed and proceeds along the original planned path.

[0069] When the accessibility classification indicates a restricted area, the robot reduces its travel speed to 0.3 meters per second while increasing the slip compensation control gain.

[0070] When the accessibility classification determines that the area is a restricted zone, the robot performs multi-objective path planning according to the method in step 3.

[0071] Step 205 includes the following steps 2051 to 2055:

[0072] Step 2051: Based on the conductivity value collected by the conductivity sensor, determine whether the robot is crossing the ice-water interface;

[0073] Step 2052: When it is determined that the robot is crossing the ice-water interface, freeze the current ice surface SLAM factor map and record the robot's last pose at the current moment. ;

[0074] Step 2053: Activate the visual-inertial-sonar tightly coupled SLAM system in underwater mode, using the final pose of the ice surface SLAM recorded in step 2052. As the initial pose of the first frame in underwater SLAM, it ensures the continuity of pose before and after the switch.

[0075] Step 2054: Align the coordinate system of the ice surface SLAM with the coordinate system of the underwater SLAM using the SE(3) rigid body transformation.

[0076] Step 2055: After completing coordinate system alignment, restore the normal state estimation of underwater SLAM, and the robot continues to move forward.

[0077] Furthermore, step 3 includes steps 301 to 305:

[0078] Step 301: The multi-objective path planning model establishes a global map;

[0079] Step 302: Define the multi-objective function of the multi-objective path planning model;

[0080] The multi-objective function is a weighted sum of four costs, calculated using the following formula:

[0081] Cost = α · L + β · E + γ · R + δ · T;

[0082] Where: L is the path length; E is the estimated energy consumption; R is the ice risk; T is the estimated time in seconds, T = L ÷ v, v = ·(1-0.5·s), s is the maximum speed; s is the slip ratio;

[0083] Step 303: The multi-objective path planning model performs global path planning and generates macro-paths on the global map;

[0084] Step 304: The multi-objective path planning model performs local path planning;

[0085] Step 305: The multi-objective path planning model uses vision-inertial-wheel speed tightly coupled SLAM for localization to achieve real-time pose optimization.

[0086] The ice-water amphibious robot and its control method provided by this invention have the following advantages compared with the prior art:

[0087] 1. The front guide arm integrates obstacle crossing, ice-crossing guidance, and active center of gravity adjustment, forming a master-slave synergy with the rear wheel propulsion mechanism, thus solving the problem of poor structural coordination. Precise pitch angle control is achieved by driving the guide arm through a stepper motor and flange, and the guide arm angle can be actively adjusted according to the ice slope and the operational stage, solving the problem of the lack of an active control mechanism in the front of existing robots. The propeller is driven by a servo motor, allowing for rapid switching between wheel and paddle-type propulsion. The structure is compact, the switching is reliable, and it adapts to the different power requirements of ice surface propulsion and underwater propulsion.

[0088] 2. Thin-film pressure sensors are installed on all four axles to collect real-time data on the wheel-to-ground contact force distribution. The guide arm acts as an active counterweight mechanism, swinging to adjust the center of gravity for real-time calculation and active adjustment. A millimeter-wave radar and ultrasonic thickness sensor are integrated at the front of the base plate to detect the thickness and internal structure of the ice layer in real time, providing sensing data for path planning and ice-entry decisions. Conductivity sensors are integrated on both sides of the base plate to determine the states of the ice surface, interface, and underwater media, enabling autonomous switching between propulsion modes and the SLAM framework.

[0089] 3. The robot is reinforced with an aluminum alloy sealing box and an aluminum profile base plate, which has strong resistance to ice surface impact; the buoyancy foam is attached to give the robot a slight positive buoyancy and stable posture in water.

[0090] 4. Predictive center of gravity adjustment is achieved through the LSTM overturning prediction model instead of passive stabilization. Active center of gravity adjustment and torque distribution are executed, which significantly improves the safety during the transition between ice and water, enhances the anti-overturning capability, and solves the problem of overturning control lag. By constructing an ice surface slip classification model and adopting differentiated compensation strategies such as pulse propulsion, differential speed reversal, and early mode switching, the problem of single slip control is solved. Seamless switching of dual-medium SLAM and alignment with the SE(3) rigid body transformation coordinate system ensure the consistency of positioning between ice surface and underwater operations, and solves the problem of discontinuous positioning across media. For the first time, ice layer bearing capacity assessment-driven path planning is introduced. By establishing an ice layer bearing capacity model and adding an ice layer risk penalty term to the improved A* algorithm, thin ice areas can be avoided autonomously, reducing operational risks and solving the problem of insufficient safety in path planning. Attached Figure Description

[0091] Figure 1 This is a structural schematic diagram of the ice-water amphibious robot provided by the present invention;

[0092] Figure 2 This is a schematic diagram of the skeleton of the ice and water amphibious robot provided by the present invention;

[0093] Figure 3This is a top view of the skeleton of the ice and water amphibious robot provided by the present invention;

[0094] Figure 4 This is a bottom view of the skeleton of the ice-water amphibious robot provided by the present invention;

[0095] Figure 5 This is a schematic diagram of the paddle wheel switching provided by the present invention;

[0096] Figure 6 This is a schematic diagram of the sensor arrangement of the ice-water amphibious robot provided by the present invention;

[0097] Figure 7 This is a schematic diagram of the ice-water amphibious robot overcoming obstacles provided by the present invention;

[0098] Figure 8 This is a schematic diagram of the ice-water amphibious robot provided by the present invention entering the ice;

[0099] Figure 9 This is a flowchart of the control method for the ice-water amphibious robot provided by the present invention;

[0100] Figure 10 This is a flowchart of step 2 provided by the present invention;

[0101] Figure 11 This is a flowchart of step 3 provided by the present invention.

[0102] Reference numerals: 1. First drive motor; 2. Guide arm; 3. Flange; 4. Servo motor; 5. Second drive motor; 6. Paddle wheel; 7. Square box; 8. Small box; 9. Reinforcing profile; 10. Buoyancy foam; 11. Binocular vision camera; 12. Millimeter-wave radar; 13. Conductivity sensor; 14. Ice layer acoustic thickness sensor; 15. Temperature and humidity sensor; 16. Base plate; 17. Housing; 18. Ice. Detailed Implementation

[0103] To make the technical problems solved by this invention, the technical solutions adopted, and the technical effects achieved clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings, not all of them.

[0104] Example 1

[0105] like Figure 1-4 As shown in the figure, an amphibious robot provided by an embodiment of the present invention includes: a base plate 16, a first drive motor 1, a flange 3, a guide arm 2, a servo motor 4, and a propeller wheel 6.

[0106] The base plate 16 has a rectangular opening at the center of its front portion, and a first drive motor 1 is mounted on one side of the rectangular opening. The output shaft of the first drive motor 1 is connected to the input end of the flange 3; the flange 3 is located above the rectangular opening of the base plate 16; the output end of the flange 3 is connected to the guide arm 2, specifically to the main rod of the guide arm 2; the guide arm 2 is T-shaped.

[0107] Servo motors 4 are respectively installed at both ends of the rear portion of the base plate 16 and both ends of the guide arm 2. The servo motors 4 are used for switching between propeller and wheel states. The servo motors 4 can be mg996r servo motors. A second motor 5 is installed on the servo motor 4. The second motor 5 can be a C6374 motor. A propeller wheel 6 is installed at the output end of the second motor 5. The propeller wheel 6 has multiple evenly arranged blades, and the edge of the propeller wheel 6 is embedded with blades to ensure friction when traveling on ice 18. The propeller wheel 6 of this invention functions as a propeller in water and as a wheel on ice 18. Figure 5 As shown, when the initial angle of the servo motor 4 is 0°, the axis of the propeller wheel 6 is perpendicular to the axis of the robot body, and the propeller wheel 6 contacts the ice surface as a wheel (wheel mode); when the servo motor 4 rotates 90°, the axis of the propeller wheel 6 is parallel to the axis of the robot body, generating horizontal thrust (propeller mode); the switching time of the propeller wheel 6 is about 0.5 seconds.

[0108] A housing 17 is fitted above the base plate 16; the housing 17 has elongated holes at positions corresponding to the guide arm 2; buoyancy foam 10 is adhered to the outer surfaces of the base plate 16 and the housing 17, and the buoyancy foam 10 is adhered to the exposed parts of the body to increase buoyancy. Reinforcing profiles 9 are provided at multiple locations on the lower surface of the base plate 16 to reinforce the base plate 16.

[0109] A square box 7 is installed at the rear center of the base plate 16, and the battery and main controller are installed inside the square box 7. A small box 8 is installed at the center of the lower surface of the base plate 16, and a gyroscope and a water pressure sensor are installed inside the small box 8. The gyroscope is an MPU6050 gyroscope. The small box 8 is installed parallel to the base plate 16 to ensure accurate attitude monitoring.

[0110] In this embodiment, as Figure 6 As shown, a binocular vision camera 11, a millimeter-wave radar 12, and an ice layer acoustic thickness sensor 14 are installed at the front of the housing 17; conductivity sensors 13 are installed on both sides of the base plate 16; a temperature and humidity sensor 15 is installed on the housing 17; and a wheel axle contact sensor is installed on the propeller wheel 6.

[0111] The amphibious robot of this invention has the functions of underwater walking, obstacle crossing, ice climbing, and ice traversing. When walking underwater, the four propeller wheels 6 act as propellers, propelling the robot underwater. Figure 7As shown, when overcoming obstacles, the guide arm 2 is lifted and placed on the surface of ice 18. The rear propeller wheel 6 propels the robot forward in a paddle-like motion. When the robot generates an upward tilt angle, the guide arm 2 swings in coordination with the rear propeller wheel 6 at a differential speed to straighten the robot body and continue crawling forward. Figure 8 As shown, when descending onto the ice, the guide arm 2 is placed under the surface of the ice 18. The rear propeller wheel 6 propels the robot forward in a paddle-like manner. When the robot reaches a downward tilt angle, it is straightened and uses buoyancy to adhere to the ice surface. After completion, the robot then transforms into a wheel-like mechanism to crawl forward. The robot's status is monitored by a gyroscope in the water flow, and the speed of the four second motors 5 is adjusted to maintain a stable posture.

[0112] The robot's ice-crossing and obstacle-crossing actions in this invention primarily rely on the first drive motor 1 to lift and lower the front T-shaped guide arm 2, while the propeller mode switching mainly relies on the rotation adjustment of the servo motor 4. The front guide arm 2 and the rear propeller wheel 6 constitute a "master-slave cooperative" system—the rear propeller wheel 6 is responsible for the main propulsion power, and the front guide arm 2 switches its working mode according to the operation stage.

[0113] Navigation mode: The propeller wheel 6 is converted into a propeller form by the servo motor 4, and the front and rear propeller wheels 6 are paddle type; the robot is propelled to navigate in water, and the front propeller wheel 6 is controlled by the propeller to control the robot's navigation direction.

[0114] Obstacle crossing guidance mode: the front paddle wheel 6 is wheel type, and the rear paddle wheel 6 is paddle type; the guide arm 2 is raised upward, and the front paddle wheel 2 contacts the ice surface at low speed as an auxiliary support point to assist the rear paddle wheel 6 in climbing the ice surface.

[0115] Ice-entry guidance mode: the front propeller wheel 6 is wheel type, and the rear propeller wheel 6 is paddle type; the guide arm 2 swings down below the ice surface, and the front propeller wheel 6 propels in a paddle type to guide the robot to generate a downward angle to complete the water entry;

[0116] Ice walking mode: All propeller wheels 6 are converted into wheels by the servo motor 4. The front and rear propeller wheels 6 are wheeled, and the shape is like a car walking on the ice.

[0117] Center of gravity adjustment mode: The guide arm 2 remains horizontal, the front paddle wheel 2 stops rotating, and actively swings according to the robot's posture to adjust the overall center of gravity and achieve anti-tipping.

[0118] Specifically, after the base plate 16 is fitted with the housing 17, the overall dimensions are 450mm (length) × 320mm (width) × 150mm (height), with a total weight (including battery) of approximately 4.2kg and a maximum designed operating water depth of 50m. The base plate 16 is made of 6061 aluminum alloy sheet, 3mm thick, and has a rectangular shape with rounded corners. Two 20mm × 20mm × L400mm reinforcing profiles 9 are installed along the length of the lower surface of the base plate 16. These aluminum profiles are fixed with M3 × 8mm countersunk screws at 50mm intervals, forming a composite reinforced structure with the base plate 16 to effectively resist the impact of ice and underwater pressure. The buoyancy foam 10 is made of closed-cell polyethylene foam with a density of 0.05g / cm³, providing a total buoyancy of approximately 2.5kg, giving the robot a slight positive buoyancy in water (buoyancy approximately 1.1 times its weight). The first drive motor 1 uses a 57BYGH56 motor with a transmission ratio of 100:1, used for the swinging of the guide arm 2. The binocular vision camera 11 uses an Intel RealSense D435 for obstacle recognition and depth estimation. The millimeter-wave radar 12 uses an IWR6843 for ice thickness detection and underwater terrain surveying. The conductivity sensor 13 uses an InPro7000 for real-time determination of the interface between the ice surface and the underwater environment. The ice acoustic thickness sensor 14 uses a Geotech LD-1200 for accurate ice thickness measurement. The temperature and humidity sensor 15 uses an SHT31 for assessing icing risk. The wheel axle contact force sensor uses an FSR-406 for monitoring ice surface contact force distribution. The water pressure sensor uses an MS5837 for underwater depth determination and buoyancy control. The upper-level computing module of the main controller uses a Raspberry Pi 4B to run visual SLAM and reinforcement learning inference. The lower-level operation control module of the main controller uses an STM32F4 for sensor acquisition, motor drive, and attitude closed-loop control.

[0119] Example 2

[0120] Figure 9 As shown, an embodiment of the present invention provides a control method for an amphibious robot based on any embodiment of the present invention, comprising the following processes:

[0121] Step 1: Real-time acquisition of robot operational and environmental data. The operational data includes pitch angle, roll angle, force distribution across the four axles, wheel speed, three-axis acceleration, three-axis angular velocity, and body depth. The environmental data includes forward depth map, ice thickness, internal ice structure, conductivity, ambient temperature, and ambient humidity.

[0122] Specifically, the system collects pitch angle, roll angle, three-axis acceleration, and three-axis angular velocity using a gyroscope; collects the force distribution of the four wheel axles using a wheel axle force sensor; measures wheel speed using a wheel speed encoder integrated with the second motor; measures the fuselage depth using a water pressure sensor; acquires a forward depth map using a binocular vision camera 11; measures ice thickness using a millimeter-wave radar 12; measures the internal structure of the ice layer using an acoustic ice thickness sensor 14; measures conductivity using a conductivity sensor 13; and measures ambient temperature and humidity using a temperature and humidity sensor 15.

[0123] Step 2: Based on the real-time collected robot operation data and environmental data, determine the robot's posture stability, overturning risk level, slip type, accessibility level, and whether it crosses media. Then, based on the posture stability, overturning risk level, slip type, accessibility level, and whether it crosses media, perform robot posture control.

[0124] This step uses five models—instantaneous center of gravity position model, LSTM overturning prediction model, ice surface slip classification model, ice layer bearing capacity model, and cross-medium switching model—to determine the robot's operating status and perform attitude control.

[0125] Figure 10 As shown, step 2 includes the following steps 201 to 205:

[0126] Step 201: Determine the robot's posture stability state using the instantaneous center of gravity position model, and perform posture control on the robot based on the posture stability state.

[0127] The contact force values ​​of the four axles (F1 for the left front propeller axle, F2 for the right front propeller axle, F3 for the left rear propeller axle, and F4 for the right rear propeller axle), and the coordinates of the center of mass of guide arm 2 are recorded. , , ), the centroid coordinates of the square box 7 ( , , Input the instantaneous center of gravity position model to obtain the instantaneous center of gravity coordinates. , ), center of gravity offset (ΔX, ΔY), stability margin value Based on the center of gravity offset, stability margin value, and preset stability boundary value, the guide arm pitch angle is adjusted, the differential torque distribution of the four second motors is performed, and the body attitude is corrected in a closed loop.

[0128] The instantaneous center of gravity position model includes six mass points: the contact point of the left front rotor, the contact point of the right front rotor, the contact point of the left rear rotor, the contact point of the right rear rotor, the center of mass of the square box 7, and the center of mass of the guide arm 2. Each of the instantaneous center of gravity position models... (Usually 10ms) Collect the force sensor values ​​F1~F4 (unit: N) of the four wheel axles.

[0129] The instantaneous center of gravity position model is based on the known coordinates of each mass point ( , , ), calculate the instantaneous center of gravity coordinates according to the following formula:

[0130] = ;

[0131] = ;

[0132] Using the calculated instantaneous center of gravity coordinates, and based on the definition of a stable region, the actual center of gravity is always kept within the preset stable region.

[0133] The stable region is defined as | - | < (usually 20mm), and | - |< , and The preset stable center of gravity reference coordinates for the robot.

[0134] The instantaneous center of gravity position model uses the swing angle of guide arm 2 as the control variable to ensure that the actual center of gravity remains within a preset stable area. Specifically: when the calculated instantaneous center of gravity position coordinates ( , When deviating from the stable region, the instantaneous center of gravity position model is based on the center of gravity offset (ΔX, ΔY) and the stability margin value. Calculate the required swing angle Δ of guide arm 2. The swing angle Δ of guide arm 2 It is positively correlated with the center of gravity offset (ΔX, ΔY); the first drive motor 1 swings according to the guide arm 2's swing angle Δ The drive guide arm 2 swings in the opposite direction of the center of gravity shift, pulling the machine's center of gravity back to the stable region. Simultaneously, based on the center of gravity shift and stability margin, differential torque distribution is applied to the second motors 5 connected to the four axles, reducing the torque of the second motor 5 on the side of the center of gravity shift and correspondingly increasing the torque of the second motor 5 on the non-shifted side. The above adjustment process is based on... (Usually using 10ms) as the cycle for cyclic execution, forming a closed-loop correction control of the robot's attitude, ensuring that the robot maintains stable attitude during movement on ice, obstacle crossing, going down ice and going up ice.

[0135] Step 202: Determine the robot's overturning risk level using the LSTM overturning prediction model, and control the robot's posture based on the overturning risk level.

[0136] The LSTM overturning prediction model employs a four-layer network structure: the first layer is an LSTM recurrent neural network layer used to extract time-series features, with 10-dimensional input and 32-dimensional output; the second layer is a Dropout layer with a dropout rate of 0.2 to prevent overfitting; the third layer is a fully connected layer containing 16 neurons and using the ReLU activation function; and the fourth layer is an output layer containing one neuron and using the Sigmoid activation function to output the overturning probability. Overturning direction Estimated time to capsize The LSTM overturning prediction model was built in the Gazebo simulation environment using an ice-water physics model to simulate overturning scenarios under different combinations of pitch angle, roll angle, wheel contact force, and wheel speed. 100,000 sets of data were collected as training samples. Labels indicating whether an overturning would occur within the next 0.3 seconds were used. The model was trained using the Adam optimizer with a learning rate of 0.001, achieving an accuracy of over 95% on the validation set.

[0137] The input to the LSTM overturning prediction model is the past Pitch and roll angles sampled within a time period Four axle contact force values ​​F1~F4 and four wheel speeds ω1~ω4 were collected. The time interval for data acquisition was... . For example, 100ms. The input sequence has a length of 10 time steps (100ms each) and a total of 10-dimensional features.

[0138] The output of the LSTM overturning prediction model: Future (Typically, the overturning probability is used within 0.3 seconds) Overturning direction Estimated time to capsize . This represents the probability of overturning, with the overturning probability threshold set to 'a' (usually 0.7).

[0139] The attitude control of the LSTM overturning prediction model: when the overturning probability When the overturning probability threshold 'a' is reached, guide arm 2 swings in the opposite direction of overturning. (Typically 15°~30°), while reducing the torque of the second motor on the overturning side by k% (typically 30%), thereby increasing the thrust on the non-overturning side and adjusting the dynamic center of gravity of the fuselage.

[0140] This step uses the LSTM overturning prediction model to predict the future through deep learning. By predicting the probability of overturning within a given timeframe, and by proactively adjusting the center of gravity or redistributing propulsion, anti-overturning control can be achieved in a predictive rather than reactive manner.

[0141] Step 203: Determine the robot's slip type using the ice surface slip classification model, and control the robot's posture based on the slip type.

[0142] This step inputs the wheel speed change rate Δω, fuselage displacement change rate Δv, slip rate s, roll angle change rate Δφ, and electrical conductivity EC into the ice surface slip classification model to obtain the slip type (Type: Type I / Type II / Type III / Normal) and slip severity. , Recover confidence value The robot's attitude is controlled based on the slip type, slip severity, recovery confidence, current wheel speed, and current guide arm angle. The attitude control based on the slip type includes pulse propulsion (Type I), differential reversal + guide arm lateral swing compensation (Type II), advance mode switching and guide arm swing down (Type III) or conventional PI control (normal).

[0143] The ice surface slip classification model categorizes slip into three types based on sudden increases in wheel speed, lateral displacement of the fuselage, and abrupt changes in electrical conductivity: acceleration slip (Type I), lateral slip (Type II), and ice-water interface slip (Type III). Differentiated attitude control methods, such as pulse propulsion, differential reversal, guide arm lateral swing compensation, and advance mode switching, are employed for each type.

[0144] Step 203 includes steps 2031 to 2033:

[0145] Step 2031, the ice surface slip classification model calculates each The slip detection parameters for the time interval include wheel speed change rate Δω, fuselage displacement change rate Δv, slip rate s, roll angle change rate Δφ, electrical conductivity EC, and left and right wheel speed difference. For example, 20ms.

[0146] The formula for calculating the wheel speed change rate Δω:

[0147] Δω = ;

[0148] in, The wheel speed at the current moment, in rpm; Δt represents the wheel speed at the previous moment, in rpm; Δt represents the sampling time interval, in seconds.

[0149] Formula for calculating the rate of change of fuselage displacement Δv:

[0150] Δv = ∫ a(t) dt;

[0151] Where a(t) is the fuselage acceleration measured by the IMU, in m / s²; dt is the integration time step, in s.

[0152] The formula for calculating slip ratio s:

[0153] s = ;

[0154] The slip ratio s is dimensionless and ranges from 0 to 1.

[0155] The formula for calculating the roll angle change rate Δφ is as follows:

[0156] Δφ = ;

[0157] in, The current roll angle, in degrees. Δt is the roll angle at the previous moment, in °; Δt is the sampling time interval, in s; Δφ is in ° / s.

[0158] The conductivity EC is measured in μS / cm and is collected in real time by four conductivity sensors 13 on both sides of the base plate 16.

[0159] Formula for calculating the speed difference between the left and right wheels:

[0160] | - | = | |;

[0161] in, Indicates the rotational speed of the two wheels on the left; Indicates the rotational speed of the two wheels on the right side; , , , These are the wheel speeds of the left front rotor 6, the right front rotor 6, the left rear rotor 6, and the right rear rotor 6, respectively, in rpm.

[0162] Step 2032: Determine the type of ice surface slippage based on the calculated slippage detection parameters.

[0163] When the slip ratio s > the first slip ratio threshold (e.g., 0.3) and the fuselage displacement change rate Δv < the first fuselage displacement change rate threshold (e.g., 0.1), the slip type is determined to be acceleration slip (Type I); when the fuselage displacement change rate Δv > the second fuselage displacement change rate threshold (e.g., 20° / s) and the difference in speed between the left and right wheels | - When the left and right wheel speed difference threshold (the left and right wheel speed difference threshold is, for example, 50 rpm) is reached, the slip type is determined to be lateral slip (Type II); when the conductivity EC changes abruptly within the range of 100~1000 μS / cm (the rate of change of conductivity between two adjacent samples is greater than the rate of change of conductivity threshold, the rate of change of conductivity threshold is, for example, 50 μS / cm / 10ms) and the slip rate s is greater than the second slip rate threshold (the first slip rate is, for example, 0.2), the slip type is determined to be ice-water interface slip (Type III); when none of the above conditions are met, the slip type is determined to be normal driving.

[0164] Step 2033: If the slip type is determined to be acceleration slip (Type I), pulse propulsion is used; if the slip type is determined to be lateral slip (Type II), differential reversal and guide arm 2 side swing are used; if the slip type is determined to be ice-water interface slip (Type III), advance switching propulsion mode and guide arm downward swing are used; if the slip type is determined to be normal driving, conventional PI control is used.

[0165] The pulsed propulsion is: in a periodic... Intermittent high torque output mode with a duty cycle of s% (usually 0.5 seconds) and a peak torque of 200% of the normal torque, generates intermittent impact force to break the water film on the ice surface, restore wheel grip, until the slip ratio s drops below 0.1.

[0166] The differential reversal and guide arm 2 lateral swing are as follows: the paddle wheel 6 on the overturning side reverses and applies a reverse driving force of 20% of the rated torque, while the outer wheel maintains forward propulsion; simultaneously, the guide arm lateral swings towards the overturning side. (Usually 15°), increase the support width and increase the anti-rollover moment until the roll angle change rate Δφ drops below 10° / s.

[0167] The aforementioned advance mode switching and guide arm lowering are as follows: when the conductivity EC detects the ice-water interface, before the rear wheel mode completely fails, the advance mode switching and guide arm lowering are initiated. (Typically 0.5 seconds) Switch the rear propeller from wheel mode to propeller mode; simultaneously, guide arm 2 swings downwards. (Usually 10°) guides the fuselage to generate a downward angle, smoothly transitioning to underwater propulsion mode and avoiding loss of control due to interface slippage.

[0168] The conventional PI control is a proportional-integral control based on the slip ratio s error, with the control law being τ = · + ·∫ ·dt, where The target slip ratio is the difference between the target slip ratio and the current slip ratio, with the target slip ratio set as b (usually 0.1). and These are the tuned control parameters.

[0169] Step 204: Determine the robot's accessibility level using the ice layer bearing capacity model, and control the robot's posture based on the accessibility level.

[0170] The ice thickness detected by millimeter-wave radar 12, the temperature and humidity data collected by temperature and humidity sensor 15, and the ice thickness distribution measured by ice acoustic thickness sensor 14 are input into the ice bearing capacity model to obtain the ice bearing capacity value, passability cost, and safety level value. Based on the ice bearing capacity value, passability cost, safety level, current robot position, and target point position value, global path replanning, travel speed limit, and ice entry condition judgment are performed.

[0171] Step 204 includes the following steps 2041 to 2044:

[0172] Step 2041: The ice layer bearing capacity model determines the thickness h of the ice layer in front based on the echo signal detected by the millimeter-wave radar 12 and the pulse reflection time of the ice layer acoustic thickness sensor 14.

[0173] Specifically, millimeter-wave radar 12 each (Typically 100 milliseconds) Outputs the thickness of the ice layer ahead, h, in millimeters, with a detection range of [missing information]. (0.5 m to 5 m). The ice layer acoustic thickness sensor 14 emits a 200 kHz ultrasonic pulse and receives the echo from the bottom of the ice layer. It measures the time difference between pulse emission and reception, and calculates the ice layer thickness by combining the sound velocity in the ice layer, as a supplementary verification of the detection value of the millimeter-wave radar 12.

[0174] Step 2042: The ice layer bearing capacity model calculates the ice layer bearing capacity based on the temperature data T (unit: degrees Celsius) and humidity data RH (unit: percentage) collected in real time by the temperature and humidity sensor 15, combined with the ice layer thickness h.

[0175] The formula for calculating the bearing capacity of the ice layer is:

[0176] = k · h · (1 + 0.02 · (T - (-10)));

[0177] in, The formula represents the ice layer's load-bearing capacity in Newtons (N); k is a calibration coefficient of 0.5 Newtons per millimeter (N / mm); h is the ice layer thickness in millimeters; T is the ambient temperature in degrees Celsius (°C); and -10 is the reference temperature, i.e., minus 10 degrees Celsius. The physical meaning of this formula is that the ice layer's load-bearing capacity is directly proportional to its thickness, and for every 1 degree Celsius increase in temperature (i.e., the ice layer warms up), the load-bearing capacity increases by approximately 2%; for every 1 degree Celsius decrease in temperature, the load-bearing capacity decreases by approximately 2%.

[0178] Step 2043, the ice layer bearing capacity model is based on the calculated ice layer bearing capacity. The passability classification is determined based on the ice thickness h; the passability classification includes three levels: safe zone, warning zone, and restricted zone.

[0179] When the ice thickness h is greater than 100 mm or the ice bearing capacity When the load is greater than 200 Newtons, the traversability classification is safe zone; when the ice thickness h is greater than 50 mm and less than or equal to 100 mm, or the ice bearing capacity... For ice loads greater than 100 Newtons and less than or equal to 200 Newtons, the trafficability classification is a warning zone; when the ice thickness h is less than or equal to 50 mm or the ice bearing capacity... When the power is less than or equal to 100 Newtons, the accessibility level is restricted.

[0180] A safe zone indicates that the ice is strong enough for the robot to pass safely at normal speed. A warning zone indicates that the ice has limited load-bearing capacity, and the robot needs to reduce its speed and increase its slip compensation gain before proceeding cautiously. A restricted zone indicates that the ice is too thin or has insufficient load-bearing capacity, and the robot cannot pass safely and must detour.

[0181] Step 2044: The ice layer bearing capacity model performs robot posture control based on the determined accessibility classification.

[0182] When the accessibility level determines that the area is safe, the robot maintains its normal travel speed (maximum 0.5 meters per second) and proceeds along the originally planned path.

[0183] When the accessibility classification indicates a restricted zone, the robot reduces its travel speed to 0.3 meters per second while simultaneously increasing the slip compensation control gain (increasing the proportional gain of the slip controller). Increase by 50% to enhance the ability to suppress slippage on the ice surface, and prioritize routes with thicker ice or higher load-bearing capacity.

[0184] When the accessibility classification determines that the area is a restricted area, the robot performs multi-objective path planning according to the method in step 3.

[0185] The improved A* algorithm is invoked for multi-objective path planning to avoid all restricted areas; at the same time, ice danger warning information is sent to the ground station.

[0186] Step 205: Determine whether the robot is crossing media using the cross-media switching model, and control the robot's posture based on whether it is crossing media.

[0187] Step 205 includes the following steps 2051 to 2055:

[0188] Step 2051: Based on the conductivity value collected by conductivity sensor 13, determine whether the robot is crossing the ice-water interface.

[0189] When the conductivity value jumps from below 100 μS / cm to above 500 μS / cm within 10 milliseconds, it indicates that the robot is crossing the ice-water interface. Specifically, when the robot is on ice, the conductivity sensor measures a conductivity value of less than 100 μS / cm; when the robot is underwater, the conductivity sensor measures a conductivity value of more than 500 μS / cm. When the conductivity value jumps from below 100 μS / cm to above 500 μS / cm within 10 milliseconds, it indicates that the robot is crossing the ice-water interface, triggering the cross-medium switching process.

[0190] Step 2052: When it is determined that the robot is crossing the ice-water interface, freeze the current ice surface SLAM factor map and record the robot's last pose at the current moment. .

[0191] The factor graph state of the vision-inertial-wheel speed tightly coupled SLAM system operating in ice surface mode is frozen and saved, and the robot's last pose at the current moment is recorded. (Including position and attitude information).

[0192] The described vision-inertial-wheel speed tightly coupled SLAM system is a localization and mapping method that deeply fuses raw observation data from three sensors—a binocular vision camera, an IMU (Integrated Measurement Unit), and a wheel speed encoder—within the same optimization framework. Specifically, the binocular vision camera acquires images at a frequency of 30Hz and extracts ORB feature points (500 per frame), providing visual reprojection constraints; the IMU acquires acceleration and angular velocity at a frequency of 100Hz and pre-integrates them, providing short-term motion constraints; and the wheel speed encoder acquires four-wheel rotational speeds at a frequency of 100Hz and pre-integrates them, providing wheel odometry constraints. These three constraints are uniformly constructed into a factor graph, and real-time nonlinear optimization is performed using the iSAM2 incremental smoothing algorithm to solve for the optimal robot pose. This method achieves high-precision and robust localization through sensor complementarity: vision provides rich environmental features, the IMU provides high-frequency dynamic response, and wheel speed provides drift-free displacement constraints. When ice surface features are scarce, the system automatically increases the weight of wheel speed pre-integration; when slippage is detected, the confidence of wheel speed pre-integration is reduced, thereby ensuring continuous and stable robot localization in ice environments.

[0193] Step 2053: Activate the visual-inertial-sonar tightly coupled SLAM system in underwater mode, using the final pose of the ice surface SLAM recorded in step 2052. As the initial pose of the first frame in underwater SLAM, it ensures the continuity of pose before and after the switch.

[0194] Step 2054: Align the coordinate systems of the ice surface SLAM with those of the underwater SLAM using the SE(3) rigid body transformation.

[0195] SE(3) is a rigid body transformation group in three-dimensional space, containing two components: rotation and translation. The specific alignment method is as follows: = · ,in This represents the first and second poses during the initialization of the underwater SLAM system. This indicates the final pose of SLAM on the ice surface; This is the coordinate transformation matrix obtained through least squares matching calculation.

[0196] Step 2055: After completing coordinate system alignment, restore the normal state estimation of underwater SLAM, and the robot continues to move forward.

[0197] The entire cross-medium switching process takes less than 100 milliseconds, ensuring the continuity and stability of the robot's positioning information when crossing the ice-water interface.

[0198] Step 3: Based on the real-time collected environmental data of the robot, perform multi-objective path planning using a multi-objective path planning model.

[0199] Figure 11 As shown, step 3 includes steps 301 to 305:

[0200] Step 301: The multi-objective path planning model establishes a global map.

[0201] The robot creates a 10-meter by 10-meter grid map, with each grid having a resolution of 0.1 meters. The global map contains ice accessibility cost information (from the ice bearing capacity model in step 204) and obstacle information (from the binocular vision camera 11 and millimeter-wave radar 12).

[0202] Step 302: Define the multi-objective function of the multi-objective path planning model.

[0203] The multi-objective function is a weighted sum of four costs, calculated using the following formula:

[0204] Cost = α · L + β · E + γ · R + δ · T;

[0205] Where: L is the path length in meters, representing the total distance traveled from the starting point to the end point; E is the estimated energy consumption in ampere-hours (Ah), calculated based on the path gradient and distance; R is the ice risk, with a value of 100 for the restricted area, 50 for the warning area, and 0 for the safe area; T is the estimated time in seconds, T = L ÷ v, where v = ·(1-0.5·s), where The maximum speed is 0.5 meters per second; s is the slip ratio, and the default weight values ​​are: α=0.4, β=0.2, γ=0.3, δ=0.1, which can be dynamically adjusted according to task requirements (e.g., increase the weight of β when energy saving is prioritized, and increase the weight of γ when safety is prioritized).

[0206] Step 303: The multi-objective path planning model performs global path planning and generates macro-paths on the global map.

[0207] The multi-objective path planning model uses an improved A* algorithm to generate macroscopic paths on a global map. The A* algorithm is a heuristic path search algorithm that finds the optimal path by comprehensively considering both the actual cost from the starting point to the current node and the estimated cost from the current node to the target node. This invention improves the A* algorithm by adding risk penalties, enabling it to automatically prioritize safe zone paths, avoid warning zones as much as possible, and completely avoid restricted areas.

[0208] The heuristic function h(n) of the improved A* algorithm is: h(n) = Euclidean distance + ice risk penalty. The ice risk penalty is set to: 100 for passing through the restricted area, 50 for passing through the warning area, and 0 for passing through the safe area.

[0209] Step 304: The multi-objective path planning model performs local path planning.

[0210] After the global macroscopic path is generated, the robot performs local real-time obstacle avoidance using the Dynamic Window Method (DWA) every 200 milliseconds. The velocity window of the Dynamic Window Method is set to a linear velocity v within the range of 0 to 0.5 meters per second, and an angular velocity ω within the range of -30 degrees per second to +30 degrees per second. The multi-objective path planning model searches for the optimal velocity combination within the velocity window, enabling the robot to avoid dynamically appearing obstacles while traveling along the global path.

[0211] Step 305: The multi-objective path planning model uses vision-inertial-wheel speed tightly coupled SLAM for localization to achieve real-time pose optimization.

[0212] In multi-objective path planning, the robot's localization information is provided by the SLAM system. Specifically:

[0213] When the robot is on ice, a vision-inertial-wheel speed tightly coupled SLAM is used for localization. This SLAM method extracts ORB feature points through a binocular vision camera 11, extracting 500 feature points per frame for matching and pose estimation; the wheel speed encoder of the second motor performs pre-integration every 10 milliseconds to provide short-term displacement constraints; the optimization method adopts the iSAM2 incremental smoothing algorithm based on factor graphs to achieve real-time pose optimization.

[0214] When the robot is underwater, it uses tightly coupled visual-inertial-sonar SLAM for localization. Due to poor underwater lighting conditions, visual features are replaced with edge features. Image edges are extracted using the Canny operator, and stable feature points are obtained through fast corner detection. The millimeter-wave radar 12 outputs the distance to obstacles ahead every 200 milliseconds, ranging from 0.2 meters to 5 meters. This distance information is added to the factor graph as an observation constraint to assist underwater localization. The Canny operator is an edge detection algorithm used to extract the contour boundaries of objects from images. It was proposed by John F. Canny in 1986 and is widely recognized as one of the best edge detection algorithms. In this robot, the Canny operator is used for visual feature extraction in the underwater environment, solving the problem of difficult traditional feature point extraction caused by poor underwater lighting and low contrast.

[0215] When the robot crosses the ice-water interface, the SLAM framework is switched according to the cross-medium switching process described in step 205, which will not be repeated here.

[0216] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications to the technical solutions described in the foregoing embodiments, or equivalent substitutions for some or all of the technical features, do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A control method for an ice-water amphibious robot, characterized in that, The ice and water amphibious machine includes: a base plate (16), a first drive motor (1), a flange (3), a guide arm (2), a servo motor (4), and a propeller wheel (6); The base plate (16) has a rectangular opening at the middle of its front part. A first drive motor (1) is provided on one side of the rectangular opening of the base plate (16). The output shaft of the first drive motor (1) is connected to the input end of the flange (3). The flange (3) is located above the rectangular opening of the base plate (16). The output end of the flange (3) is connected to the guide arm (2). The guide arm (2) is T-shaped. Servo motors (4) are installed at both ends of the rear part of the base plate (16) and at both ends of the guide arm (2); a second motor (5) is installed on the servo motor (4), and a propeller wheel (6) is installed at the output end of the second motor (5); the servo motor (4) can drive the propeller wheel (6) to rotate 90° to achieve the switching between wheel mode and propeller mode; A square box (7) is provided at the middle of the rear part of the base plate (16), and a battery and a main controller are installed inside the square box (7); a small box (8) is installed at the middle of the lower surface of the base plate (16), and a gyroscope and a water pressure sensor are installed inside the small box (8). The control method for the ice-water amphibious robot includes the following processes: Step 1: Collect robot's operational and environmental data in real time. The operational data includes pitch angle, roll angle, force distribution on the four axles, wheel speed, three-axis acceleration, three-axis angular velocity, and body depth. The environmental data includes forward depth map, ice thickness, internal ice structure, conductivity, ambient temperature, and ambient humidity. Step 2: Based on the real-time collected robot operation data and environmental data, determine the robot's posture stability, overturning risk level, slip type, accessibility level, and whether it crosses media, and then perform robot posture control based on the posture stability, overturning risk level, slip type, accessibility level, and whether it crosses media. Step 2 includes the following steps 201 to 205: Step 201: Determine the robot's attitude stability state using the instantaneous center of gravity position model, and perform attitude control on the robot based on the attitude stability state. In step 201, the contact force values ​​of the four wheel axles and the coordinates of the center of mass of the guide arm (2) are set. , Square box (7) centroid coordinates Input the instantaneous center of gravity position model to obtain the instantaneous center of gravity coordinates. , center of gravity offset (ΔX, ΔY), stability margin value Based on the center of gravity offset, stability margin value, and preset stability boundary value, the guide arm pitch angle is adjusted, the differential torque distribution of the four second motors (5) is performed, and the body attitude is corrected in a closed loop. Step 202: Determine the overturning risk level of the robot using the LSTM overturning prediction model, and control the robot's posture based on the overturning risk level. In step 202, the LSTM overturning prediction model adopts a four-layer network structure, including: the first layer is an LSTM recurrent neural network layer; the second layer is a Dropout layer; the third layer is a fully connected layer using the ReLU activation function; and the fourth layer is an output layer using the Sigmoid activation function. The input of the LSTM overturning prediction model is the past... Pitch and roll angles sampled within a time period 4 wheel axle contact force values ​​F1~F4, 4 wheel speeds The data collection time interval is... The output of the LSTM overturning prediction model: future Probability of overturning within a time period Overturning direction Estimated time to capsize ; The attitude control of the LSTM overturning prediction model: when the overturning probability When the overturning probability threshold a is reached, the guide arm (2) swings in the opposite direction of overturning. At the same time, reduce the torque of the second motor on the overturning side by k% Step 203: Determine the robot's slip type using the ice surface slip classification model, and control the robot's posture based on the slip type; Step 204: Determine the robot's navigability level using the ice layer bearing capacity model, and control the robot's posture based on the navigability level. Step 205: Determine whether the robot is crossing media using the cross-media switching model, and control the robot's posture based on whether it is crossing media. Step 3: Based on the real-time collected environmental data of the robot, perform multi-objective path planning using a multi-objective path planning model.

2. The control method for the ice-water amphibious robot according to claim 1, characterized in that, The propeller (6) has multiple blades evenly arranged, and the edge of the propeller (6) is embedded with blades.

3. The control method for the ice-water amphibious robot according to claim 1, characterized in that, The base plate (16) is fitted with a housing (17) above it; The housing (17) has an elongated hole at a position corresponding to the guide arm (2); The base plate (16) and the outer surface of the casing (17) are bonded with buoyancy foam (10). The bottom surface of the base plate (16) is provided with multiple reinforcing profiles (9).

4. The control method for the ice-water amphibious robot according to claim 3, characterized in that, The front of the housing (17) is equipped with a binocular vision camera (11), a millimeter-wave radar (12), and an ice layer acoustic thickness sensor (14). Conductivity sensors (13) are respectively installed on both sides of the base plate (16). A temperature and humidity sensor (15) is installed on the housing (17); A wheel axle contact force sensor is installed on the propeller wheel (6).

5. The control method for the ice-water amphibious robot according to claim 1, characterized in that, Step 203 includes steps 2031 to 2033: Step 2031, the ice surface slip classification model calculates each Slip detection parameters for time intervals, including wheel speed change rate. , fuselage displacement rate Slip ratio s, roll angle change rate Electrical conductivity (EC) and speed difference between left and right wheels; The formula for calculating the wheel speed change rate Δω: ； in, This represents the wheel speed at the current moment. The wheel speed is the speed at the previous moment; Δt is the sampling time interval. Formula for calculating the rate of change of fuselage displacement Δv: ； Where a(t) is the fuselage acceleration measured by the IMU; dt is the integration time step; The formula for calculating slip ratio s: ； Wherein, the slip ratio s is dimensionless and ranges from 0 to 1; The formula for calculating the roll angle change rate Δφ is as follows: ； in, The current roll angle; The roll angle at the previous moment; The conductivity EC is collected in real time by four conductivity sensors (13) on both sides of the base plate (16); Formula for calculating the speed difference between the left and right wheels: ； in, Indicates the rotational speed of the two wheels on the left; Indicates the rotational speed of the two wheels on the right side; The speeds of the left front rotor (6), the right front rotor (6), the left rear rotor (6), and the right rear rotor (6) are respectively. Step 2032: Determine the type of ice surface slippage based on the calculated slippage detection parameters; When the slip ratio s > the first slip ratio threshold and the fuselage displacement change rate Δv < the first fuselage displacement change rate threshold, the slip type is determined to be acceleration slip; when the fuselage displacement change rate Δv > the second fuselage displacement change rate threshold and the speed difference between the left and right wheels is greater than the threshold, the slip type is determined to be acceleration slip. When the speed difference between the left and right wheels reaches the threshold, the slip type is determined to be lateral slip; when the conductivity EC changes abruptly within the range of 100~1000 μS / cm and the slip rate s > the second slip rate threshold, the slip type is determined to be ice-water interface slip; when none of the above conditions are met, the slip type is determined to be normal driving. Step 2033: When the slip type is determined to be acceleration slip, pulse propulsion is adopted; when the slip type is determined to be lateral slip, differential reversal and guide arm (2) side swing are adopted; when the slip type is determined to be ice-water interface slip, advance switching propulsion mode and guide arm downward swing are adopted; when the slip type is determined to be normal driving, conventional PI control is adopted.

6. The control method for the ice-water amphibious robot according to claim 4, characterized in that, Step 204 includes the following steps 2041 to 2044: Step 2041, the ice layer bearing capacity model determines the thickness h of the ice layer in front based on the echo signal detected by the millimeter-wave radar (12) and the pulse reflection time of the ice layer acoustic thickness sensor (14); Step 2042, the ice layer bearing capacity model calculates the ice layer bearing capacity based on the temperature data T and humidity data RH collected in real time by the temperature and humidity sensor (15) and the ice layer thickness h; The formula for calculating the bearing capacity of the ice layer is: ； in, is the ice layer bearing capacity; k is the calibration coefficient; h is the ice layer thickness; T is the ambient temperature; -10 is the reference temperature; Step 2043, the ice layer bearing capacity model is based on the calculated ice layer bearing capacity. The passability classification is determined based on the ice thickness h; the passability classification includes three levels: safe zone, warning zone, and restricted zone. Step 2044: The ice layer bearing capacity model performs robot posture control based on the determined accessibility classification; When the accessibility classification determines that the area is safe, the robot maintains its normal speed and proceeds along the original planned path. When the accessibility classification indicates a restricted area, the robot reduces its travel speed to 0.3 meters per second while increasing the slip compensation control gain. When the accessibility classification determines that the area is a restricted zone, the robot performs multi-objective path planning according to the method in step 3. Step 205 includes the following steps 2051 to 2055: Step 2051: Based on the conductivity value collected by the conductivity sensor (13), determine whether the robot is crossing the ice-water interface; Step 2052: When it is determined that the robot is crossing the ice-water interface, freeze the current ice surface SLAM factor map and record the robot's last pose at the current moment. ; Step 2053: Activate the visual-inertial-sonar tightly coupled SLAM system in underwater mode, using the final pose of the ice surface SLAM recorded in step 2052. As the initial pose of the first frame in underwater SLAM, it ensures the continuity of pose before and after the switch. Step 2054: Align the coordinate system of the ice surface SLAM with the coordinate system of the underwater SLAM using the SE(3) rigid body transformation. Step 2055: After completing coordinate system alignment, restore the normal state estimation of underwater SLAM, and the robot continues to move forward.

7. The control method for the ice-water amphibious robot according to claim 1, characterized in that, Step 3 includes steps 301 to 305: Step 301: The multi-objective path planning model establishes a global map; Step 302: Define the multi-objective function of the multi-objective path planning model; The multi-objective function is a weighted sum of four costs, calculated using the following formula: ； Where: L is the path length; E is the estimated energy consumption; R is the ice layer risk; T is the estimated time, in seconds. s is the maximum speed; s is the slip ratio; Step 303: The multi-objective path planning model performs global path planning and generates macro-paths on the global map; Step 304: The multi-objective path planning model performs local path planning; Step 305: The multi-objective path planning model uses vision-inertial-wheel speed tightly coupled SLAM for localization to achieve real-time pose optimization.

Images