A hydrodynamic simulation method, electronic equipment and medium for a water-wading robot

By using equivalent geometric parameters and smooth immersion ratios, the computational efficiency and versatility issues in the hydrodynamic simulation of multi-link wading robots were resolved, achieving efficient hydrodynamic simulation and control algorithm training.

CN123116591APending Publication Date: 2026-11-17NANJING UNIV OF SCI & TECH +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611308324.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-26
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies are insufficient to effectively represent the differentiated hydrodynamic effects of multi-link wading robots in hydrodynamic simulations, and their computational efficiency and versatility are inadequate, failing to meet the parallel training requirements of reinforcement learning.

Method used

By obtaining the fixed parameters of the robot model, determining the equivalent geometric parameters and hydrodynamic application points of each link, calculating the smooth immersion ratio, synthesizing the hydrodynamic force, and transforming it to the local coordinate system to apply it to the rigid body link, hydrodynamic simulation is achieved.

Benefits of technology

A hydrodynamic simulation method that balances computational efficiency and versatility is provided, applicable to wading robots with different linkage topologies, and supports large-scale parallel training and multi-scenario control algorithm training.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123116591A_ABST
    Figure CN123116591A_ABST
Patent Text Reader

Abstract

This disclosure provides a hydrodynamic simulation method, electronic device, and storage medium for a wading robot. The method includes: determining and caching the equivalent geometric parameters and hydrodynamic application points of each link based on fixed parameters in the model; calculating the smooth submersion ratio of each link relative to the water surface at each physical step based on the dynamic parameters and equivalent geometric parameters of each link; calculating the composite hydrodynamic force of each link based on the smooth submersion ratio of each link relative to the water surface, the composite hydrodynamic force including at least buoyancy and attitude-related drag; transforming the composite hydrodynamic force from the world coordinate system to the local coordinate system of the corresponding link, and applying it to the corresponding rigid body link through an external force and external torque interface.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to the field of robot simulation and intelligent control technology, and in particular to a hydrodynamic simulation method, training method, electronic equipment and medium for a water-wading robot. Background Technology

[0002] Aquatic robots are robots that need to perform movement and operation tasks in liquid environments or water-land interface environments. Based on their operating area and mode of movement, they can be divided into: amphibious robots capable of switching between land and water, surface robots that navigate on the water surface, and underwater robots that navigate, hover, or operate underwater. All of these robots face hydrodynamic challenges in the design and training of control algorithms, including buoyancy, drag, water flow disturbances, wave disturbances, and the interaction between the propulsion mechanism and the water body. Especially for wading robots with multi-link limb structures, each link experiences differentiated hydrodynamic effects due to varying degrees of immersion, posture, and local motion states. Furthermore, the hydrodynamic coupling between links through joints poses a significant challenge to motion control.

[0003] Existing research on wading robots largely revolves around mechanism morphology design, propulsion mechanism configuration, and mode-switching control. For example, some solutions achieve water-land movement transitions through wheel-paddle combination mechanisms, fin-deformation mechanisms, or swing-arm drive mechanisms; some solutions rely on multi-sensor environmental recognition for gait switching or thruster allocation; and others apply reinforcement learning to gait control of specific amphibious robots or docking control of underwater vehicles. However, these solutions are typically targeted at specific mechanisms or control flows, focusing on gait planning, thruster allocation, sensor fusion, or mode-switching logic, and do not provide a universal, reusable hydrodynamic simulation training method for wading robots with different linkage topologies.

[0004] In the field of underwater and surface robot simulation, existing simulation platforms can support the modeling of underwater vehicles and marine robot missions, providing capabilities such as hydrostatics, hydrodynamics, propulsion, sensors, and external disturbances. These platforms are of significant value for simulating underwater vehicles, sensing systems, and marine missions. However, their hydrodynamic models and interfaces are mostly centered around the six-degree-of-freedom motion of a single rigid body, propulsion, or geometric-level collisions. When the training object is a robot with a complex multi-rigid-body structure such as legs, feet, fins, propellers, rudder surfaces, propulsion mounts, or underwater robotic arms, existing platforms struggle to directly provide link-by-link hydrodynamic modeling capabilities, failing to express the differentiated hydrodynamics generated by different attitudes and local motions of each link.

[0005] In data-driven control algorithm training, computational fluid dynamics (CFD) methods can describe complex fluid motion with high fidelity, but the computational cost is too high, failing to meet the efficiency requirements of reinforcement learning, which requires parallel processing of thousands of environments and real-time solutions per physics step. While single-rigid-body buoyancy or whole-machine damping models are computationally efficient, they lose key hydrodynamic features such as the submersion ratio of each link, attitude-related drag, and paddling directionality, making it difficult for the trained strategy to master effective underwater propulsion and entry / exit transition behaviors. Therefore, there is an urgent need for a hydrodynamic simulation training method that lies between high-fidelity CFD simulation and single-rigid-body approximation, preserving the multi-link hydrodynamic structure crucial for motion control while meeting the computational efficiency requirements of large-scale parallel training. Summary of the Invention

[0006] To address the problems in the related technologies, this disclosure provides a hydrodynamic simulation method, electronic equipment, and medium for a water-wading robot.

[0007] One aspect of this disclosure provides a hydrodynamic simulation method for a wading robot, comprising: acquiring a robot model, wherein the robot model includes fixed parameters; determining and caching the equivalent geometric parameters and hydrodynamic application points of each link based on the fixed parameters of each link in the model; calculating, at each physical step, the smooth submersion ratio of each link relative to the water surface based on the dynamic parameters of each link and the equivalent geometric parameters; calculating the composite hydrodynamic force of each link based on the smooth submersion ratio; transforming the composite hydrodynamic force from the world coordinate system to the local coordinate system of each link to form the local hydrodynamic force of the link; and applying the local hydrodynamic force of the link to the corresponding rigid body link to perform hydrodynamic simulation.

[0008] According to an embodiment of this disclosure, determining the equivalent geometric parameters of each link includes: equating each link to a regular geometric body with a uniform mass distribution; deducing the equivalent geometric dimensions based on the default mass of each link and the analytical relationship of the inertia of the regular geometric body; and calculating the equivalent volume of each link.

[0009] According to an embodiment of this disclosure, calculating the combined hydrodynamic force of each link includes: calculating the envelope range of the link in the vertical direction based on the link attitude and equivalent geometric parameters; obtaining the original submersion ratio based on the relative relationship between the water surface position and the link envelope range; performing dead zone processing and nonlinear smoothing on the original submersion ratio to obtain a smoothed submersion ratio; setting the submersion ratio to full submersion for links continuously within the water body boundary; and setting the submersion ratio to zero for links completely above the water surface; wherein the combined hydrodynamic force includes at least buoyancy and attitude-related drag.

[0010] According to the embodiments of this disclosure, the buoyancy is calculated based on the buoyancy scaling factor, water density, gravitational acceleration, equivalent volume of the connecting rod, and smooth immersion ratio, and the buoyancy direction is upward along the vertical direction of the world.

[0011] According to an embodiment of this disclosure, the attitude-related resistance includes: calculating the relative velocity of the link relative to the water flow, determining the effective upstream projection area of ​​the link in the relative velocity direction based on the link attitude, and calculating the resistance according to the velocity squared resistance model; and setting upper limits for the modulus of the relative velocity and the amplitude of the single link resistance, respectively.

[0012] According to embodiments of this disclosure, for a target link with paddling, swinging, or maneuvering functions, the synthetic hydrodynamics further includes anisotropic paddling resistance: a local axis of the link is selected as a reference axis, and the relative water flow velocity is decomposed into a tangential component along the reference axis and a normal component perpendicular to the reference axis, and the tangential resistance and normal resistance are calculated based on the tangential resistance parameters and normal resistance parameters, respectively; different types of target links are configured with different resistance parameters and gains.

[0013] According to an embodiment of this disclosure, applying local hydrodynamic forces to the corresponding rigid body link includes: numerically cleaning and limiting the combined hydrodynamic forces and stress moments of each link, and then converting them to the link's local coordinate system; setting the hydrodynamic application point of the main link as the link origin, default local centroid, or user-specified point, and setting the hydrodynamic application points of non-main links as cached default local centroids or user-configured points; and setting upper limits and absolute upper limits related to the link's mass for the combined hydrodynamic forces and stress moments, respectively.

[0014] According to embodiments of this disclosure, the configurable water surface includes at least one of a horizontal static water surface, a dynamic water surface that changes over time, a wave water surface with a spatial phase distribution, and a water flow model with a velocity field; the water training environment also includes at least one of land terrain, underwater terrain, a water-land transition slope, a water surface boundary, or an underwater operation boundary.

[0015] Another aspect of this disclosure provides a method for training a motion control algorithm for a wading robot, comprising: constructing a wading training environment with synthetic hydrodynamics in a three-dimensional physical simulation platform based on a hydrodynamic simulation method; constructing a data-driven control algorithm training task including domain randomization configuration based on the wading training environment after applying the synthetic hydrodynamics, and training a motion control strategy for the wading robot; wherein the hydrodynamic simulation method adopts any one of the hydrodynamic simulation methods described in this disclosure.

[0016] According to embodiments of this disclosure, the control algorithm training task configuration includes an observation space, an action space, a reward function, a termination condition, and a domain randomization configuration; the observation space includes at least one of robot body state, joint state, actuator state, task instructions, and environmental state estimation; the action space includes at least one of joint control variables, thruster control variables, and manipulator surface control variables; the motion control strategy does not rely on discrete motion mode labels for explicit switching and can operate continuously between land, water-land transition, water surface, and underwater scenarios.

[0017] Another aspect of this disclosure provides a hydrodynamic simulation device for a wading robot, comprising: a model acquisition unit for acquiring a robot model, wherein the robot model includes fixed parameters; an equivalent geometry buffer unit for determining and buffering the equivalent geometric parameters and hydrodynamic application points of each link based on the fixed parameters of each link in the model; an immersion ratio calculation unit for calculating, at each physical step, the smooth immersion ratio of each link relative to the water surface based on the dynamic parameters of each link and the equivalent geometric parameters; a hydrodynamic calculation unit for calculating the composite hydrodynamic force of each link based on the smooth immersion ratio, wherein the composite hydrodynamic force includes at least buoyancy and attitude-related drag; and a hydrodynamic application unit for converting the composite hydrodynamic force from the world coordinate system to the local coordinate system of each corresponding link, and applying it to the corresponding rigid body link through an external force and external torque interface.

[0018] Another aspect of this disclosure provides a training device for a motion control algorithm of a wading robot, comprising: an environment construction unit for constructing a wading training environment with synthetic hydrodynamics in a three-dimensional physical simulation platform based on a hydrodynamic simulation method; and a training task construction unit for constructing a data-driven control algorithm training task including domain randomization configuration based on the wading training environment after applying the synthetic hydrodynamics, and training a motion control strategy for the wading robot; wherein the hydrodynamic simulation method adopts any one of the hydrodynamic simulation methods described in this disclosure.

[0019] Another aspect of this disclosure provides an electronic device including: a processor; and a memory including one or more computer program instructions, wherein the one or more computer program instructions are stored in the memory and, when executed by the processor, implement any of the methods provided in this disclosure.

[0020] Another aspect of this disclosure provides a computer-readable storage medium that non-transitory stores computer-readable instructions, wherein when the computer-readable instructions are executed by a processor, they implement the method described in any of the embodiments of this disclosure.

[0021] According to the technical solutions of this disclosure, firstly, it is not limited to underwater robots, but is applicable to amphibious, surface, underwater, and underwater operation robots, covering more general control algorithm training needs; secondly, it models hydrodynamics to the robot link level, which can express the different hydrodynamic effects on different links when the immersion ratio, attitude, local velocity, and point of application are different, and is widely applicable to single-rigid-body and multi-rigid-body wading robots; furthermore, it automatically derives equivalent link geometry and local centroid, reducing the workload of manually building a dedicated hydrodynamic model for each new robot, and has strong reusability; in addition, the technical solutions of this disclosure apply buoyancy, projected area drag, and optional anisotropic paddling / manipulation drag in each physical step in the form of an external force / force rectangle, without the need for CFD-level flow field solution, which is suitable for parallel training of reinforcement learning; it can also describe land, land-water transition, surface navigation, underwater navigation, and underwater operation tasks in the same training framework, reducing the complexity of constructing dedicated control processes for different wading scenarios. Attached Figure Description

[0022] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments, taken in conjunction with the accompanying drawings. In the drawings:

[0023] Figure 1 A flowchart illustrating a hydrodynamic simulation method for a water-wading robot according to an embodiment of the present disclosure is shown schematically. Figure 2 The diagram illustrates a schematic representation of the equivalent link geometry derived from a wading robot model provided according to an embodiment of the present disclosure. Figure 3 This schematically illustrates a configurable water surface diagram of a wading robot provided according to an embodiment of the present disclosure; Figure 4 The flowchart illustrates a method for training a motion control algorithm for a wading robot according to at least one embodiment of the present disclosure. Figure 5 This schematic diagram illustrates a structural block diagram of a hydrodynamic simulation device for a wading robot according to an embodiment of the present disclosure; Figure 6 This schematic diagram illustrates a structural block diagram of a water-wading robot motion control algorithm training device according to an embodiment of the present disclosure; Figure 7 A schematic block diagram of an electronic device provided for some embodiments of this disclosure; Figure 8 A schematic block diagram of another electronic device provided for some embodiments of this disclosure; and Figure 9 This is a schematic diagram of a storage medium provided for some embodiments of this disclosure. Detailed Implementation

[0024] Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings to enable those skilled in the art to implement them more readily. Furthermore, for clarity, portions unrelated to the description of exemplary embodiments have been omitted from the drawings. In this disclosure, it should be understood that terms such as "comprising" or "having" are intended to indicate the presence of features, numbers, operations, behaviors, components, portions, or combinations thereof disclosed in this specification, and are not intended to exclude the possibility of the presence or addition of one or more other features, numbers, operations, behaviors, components, portions, or combinations thereof. It should also be noted that, without conflict, embodiments and features in the embodiments of this disclosure can be combined with each other. The present disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] The shortcomings of traditional hydrodynamic simulation methods in practical applications have become key factors restricting the training performance of control algorithms for wading robots: Firstly, there is the bottleneck of computational efficiency: although computational fluid dynamics (CFD) methods can describe complex flow fields with high fidelity, they require accurate mesh modeling of the robot's geometry, with mesh sizes typically reaching millions or even tens of millions of elements. The time consumed in a single step is too long, failing to meet the efficiency requirements of reinforcement learning, which demands parallel processing of thousands of environments and real-time solutions per physics step. Secondly, there is the contradiction between the dependence on accurate volume and simplified approximations: existing mainstream simulation platforms (such as Gazebo and NVIDIA Isaac Sim) lack built-in link-level volume immersion calculation capabilities, forcing hydrodynamic simulations to either rely on accurate CAD or... The calculation of convex hull volume or mesh volume in the model has extremely high requirements for model integrity and preprocessing. Either the robot as a whole is simplified into basic geometric shapes such as cylinders, spheres or cuboids, which cannot capture the influence of irregular hulls, multi-link limbs and dynamic water interaction on the immersion volume, resulting in significant deviations in buoyancy and drag estimation. Thirdly, it lacks versatility: fluid-structure interaction methods such as the immersion boundary method have inherent defects such as volume conservation error, step effect and ill-conditioned discrete node constraints. Moreover, existing solutions are mostly geared towards specific mechanisms or single rigid body vehicles, relying on water geometry collision or platform-specific hydrodynamic modules, which are difficult to reuse for multi-rigid body wading robots with different link topologies, resulting in poor cross-platform portability.

[0026] To address the computational efficiency bottlenecks, precision volume dependence, and lack of versatility in hydrodynamic simulation during motion control algorithm training for wading robots, this disclosure provides at least one embodiment of a hydrodynamic simulation method for wading robots. The method includes: determining and caching the equivalent geometric parameters and hydrodynamic application points of each link based on the fixed parameters of each link in the model; calculating the smooth submersion ratio of each link relative to the water surface at each physical step based on the dynamic parameters and equivalent geometric parameters of each link; calculating the composite hydrodynamic force of each link based on the smooth submersion ratio of each link relative to the water surface, wherein the composite hydrodynamic force includes at least buoyancy and attitude-related drag; transforming the composite hydrodynamic force from the world coordinate system to the local coordinate system of the corresponding link, and applying it to the corresponding rigid body link through external force and external torque interfaces.

[0027] The fixed parameters in the robot model include the default mass, default inertia, and default local centroid; the dynamic parameters of each link include the position, attitude, and water surface position of each link of the wading robot in the current physical step.

[0028] Figure 1 A flowchart illustrating a hydrodynamic simulation method for a water-wading robot provided according to an embodiment of the present disclosure is shown.

[0029] like Figure 1 As shown, the three-dimensional physical simulation platform, which has imported the multi-rigid-body mechanical model of the wading robot and constructed a wading training environment with a configurable water surface, includes operations S110-S150.

[0030] In operation S110, a robot model is obtained, wherein the robot model includes fixed parameters.

[0031] In operation S120, based on the fixed parameters of each link in the model, the equivalent geometric parameters and hydrodynamic action points of each link are determined and buffered.

[0032] In operation S130, at each physical step, the smooth submersion ratio of each link relative to the water surface is calculated based on the dynamic parameters and equivalent geometric parameters of each link.

[0033] In operation S140, the combined hydrodynamic force of each link is calculated based on the smooth immersion ratio.

[0034] In operation S150, the synthetic hydrodynamic force is transformed from the world coordinate system to the local coordinate system of each link to form the local hydrodynamic force of the link; the local hydrodynamic force of the link is applied to the corresponding rigid body link to perform hydrodynamic simulation.

[0035] The embodiments disclosed herein can overcome the shortcomings of existing hydrodynamic simulation methods, which are difficult to support large-scale parallel training due to excessive computation or dependence on precise volume. Through the intelligent modeling mechanism of equivalent geometric derivation and immersion ratio smoothing, a reasonable trade-off is achieved between fidelity and versatility. A simulation environment that takes into account both computational efficiency and multi-link hydrodynamic characteristics is stably obtained, thereby establishing a more computationally efficient and more widely adaptable general-purpose wading robot simulation platform.

[0036] For operation S110, the robot model in this embodiment includes a multi-rigid-body mechanical model of a wading robot, which can be in URDF (Unified Robot Description Format), MJCF (MuJoCo XML Format), USD (Universal Scene Description), or an equivalent multi-rigid-body description file converted from the above formats. The model includes at least the following information: the geometric and topological relationships of each rigid-body link, the joint type and parameters connecting each link, the default mass of each link, the default inertia tensor (including at least the diagonal term), the default local centroid position, actuator parameters (such as actuator type, stiffness, damping), and joint limit information. In some embodiments of this disclosure, the model may also include additional information such as the visual mesh of the links, collision geometry, sensor configuration, or thruster model. This additional information can be used to enhance the realism of the simulation or expand the types of training tasks, but does not constitute a necessary input for the hydrodynamic calculation of this invention.

[0037] In some embodiments of this disclosure, applicable water-borne robots include, but are not limited to, the following types: amphibious robots with leg structures such as quadrupedal robots, hexapod robots, and multi-legged robots; surface navigation robots such as unmanned surface vessels and unmanned boats; underwater navigation robots such as autonomous underwater vehicles (AUVs) and remotely operated underwater vehicles (ROVs); underwater operation robots with robotic arms; biomimetic underwater robots with fins or propellers; and multi-rigid-body robots with propulsion mounting structures. Although the above-mentioned robots vary in shape and operating scenarios, they all have a multi-rigid-body dynamic structure that can be decomposed into multiple rigid-body links, and therefore, the link-level hydrodynamic simulation method of this invention is applicable to all of them.

[0038] For operation S120, as mentioned above, the hydrodynamic calculations in the embodiments of this disclosure require the geometric dimensions and volumes of each link as a basis for calculating the displacement volume (buoyancy) and projected area (drag). However, robot model files typically only provide dynamic parameters such as the mass, inertia, and local centroid of the links, and may not contain regular geometric dimensions that can be directly used for hydrodynamic calculations. Even if the model contains visual meshes or collision geometry, their shapes are often complex and irregular, and directly using them for physics-step hydrodynamic calculations would incur significant computational overhead. Therefore, this invention proposes a method for inferring equivalent regular geometry based on default dynamic parameters, obtaining the geometric parameters required for hydrodynamic calculations at a lower computational cost, wherein the dynamic parameters of each link include the world coordinate position, attitude, water level, etc. of each link.

[0039] Figure 2 An example diagram is shown illustrating the equivalent link geometry derived from a wading robot model.

[0040] In some embodiments of this disclosure, the equivalent geometric parameters of each link are determined, such as... Figure 2 The following are included: The first i Each link is equivalent to a cuboid, based on the default mass of the link. and the default inertia diagonal term 、 、 ,based on

[0041]

[0042]

[0043] Solving the three equations simultaneously yields the side lengths of the equivalent cuboid. , and .

[0044] In some embodiments of this disclosure, to avoid numerical instability caused by abnormal inertia or extremely small mass, a lower limit and an upper limit can be set for the equivalent side length; equivalent volume The equivalent volume mentioned above is used for drainage volume calculation, and the default local centroid is used for the hydrodynamic application point of the leg links. This cache is preferably executed during the startup phase of the simulation event system and must precede the mass randomization and centroid randomization events. Specifically, this invention determines whether each link enables hydrodynamic calculation and which hydrodynamic model to use through link name matching, a link category table, or a user configuration file. For example, fuselage links may include links named base, trunk, base_link, etc.; leg links may include links named hip, tall, calf, foot, toe, etc.; the robotic arm links of underwater robots may include links named arm, link, gripper, etc.; and the manipulation and propulsion links of surface or underwater robots may include links named fin, rudder, propeller, thruster, etc. Through this configuration method, users can flexibly specify the link range and model parameters for hydrodynamic calculation for different types of robots, avoiding redundant calculations for links that do not require hydrodynamic action (such as purely internal transmission links), further improving simulation efficiency.

[0045] It should be noted that the equivalent cuboid is only a preferred embodiment of the present invention and does not constitute a limitation thereof. In some embodiments of the present invention, the connecting rod can also be equivalent to an ellipsoid or a cylinder, and the geometric parameters can be deduced based on the inertia analytical relationship of the corresponding geometric body. For example, for connecting rods with characteristics of revolution (such as cylindrical propeller housings or spherical joints), using an equivalent sphere or cylinder may be closer to its actual shape than a cuboid, thereby obtaining a more accurate estimate of the projected area and displacement volume. Those skilled in the art can select a suitable type of equivalent geometry based on the actual shape characteristics of the connecting rod.

[0046] While deriving equivalent geometric parameters, it is also necessary to determine and cache the hydrodynamic application points of each link. The hydrodynamic application point refers to the position where the resultant hydrodynamic force is applied on the link; its offset relative to the link's center of mass will generate an additional torque, affecting the robot's attitude response. For fuselage or main body links, since they are usually large and relatively regular in shape, the hydrodynamic application point can be set as the link origin, the default local center of mass, or a user-specified point. For non-main body links such as legs, feet, fins, propellers, rudders, thruster mounting links, or robotic arm links, the hydrodynamic application point is preferably set as the default local center of mass in the model, because the default local center of mass is usually located near the geometric center of the link, and using this as the hydrodynamic application point can better approximate the positions of the buoyancy center and the drag center. In some embodiments of this invention, users can also specify custom hydrodynamic application points for specific links through configuration files to adapt to special link shapes or hydrodynamic distributions.

[0047] In some embodiments of this disclosure, the aforementioned equivalent geometry derivation and hydrodynamic point of action caching must be performed before domain randomization. Specifically, this pre-computation is preferably performed during the simulation startup phase (e.g., during simulation environment initialization or reset events) and must precede domain randomization events such as mass randomization, centroid randomization, and actuator parameter randomization. This is because domain randomization techniques improve the robustness of training strategies to model uncertainties and real-world differences by randomly changing parameters such as the robot's mass, centroid, and actuator stiffness during training. If the equivalent geometry parameters are calculated after domain randomization, the equivalent volume and hydrodynamic point of action will change after each randomization, leading to instability in the hydrodynamic calculation baseline. This could even result in problems such as a random decrease in mass while the equivalent volume remains unchanged, or an imbalance between buoyancy and gravity. By caching and fixing the equivalent geometry parameters and hydrodynamic point of action before domain randomization, the randomization of the body mass, link mass, centroid, or actuator parameters during runtime only changes the robot's dynamic response and does not disrupt the geometric baseline used for hydrodynamic calculation, thus ensuring the consistency and stability of the hydrodynamic calculation.

[0048] For the S130 operation, the submersion ratio is the core input for subsequent buoyancy and drag calculations, and its accuracy directly affects the fidelity of the hydrodynamic simulation. For wading robots with multi-link structures, each link has different degrees of submersion due to its attitude and position, and the submersion ratio changes rapidly during entry and exit from the water. If the original submersion ratio obtained by linear interpolation is used directly, hydrodynamic abrupt changes will occur when the link crosses the free surface, leading to numerical instability and strategy oscillations during training. Therefore, this invention obtains a continuous and stable smooth submersion ratio through three steps: envelope height calculation, original submersion ratio estimation, and nonlinear smoothing.

[0049] In some embodiments of this disclosure, calculating the smooth submersion ratio of each link relative to the water surface includes: according to the first... i The attitude rotation matrix of each link and equivalent side length , and Calculate the half-height of the link in the vertical direction worldwide. .

[0050] ; in, , and These are the three elements in the third row of the attitude rotation matrix, representing the projection components of the link's local coordinate axes onto the world vertical direction. The physical meaning of this formula is that the maximum half-height of the cuboid's envelope in the world vertical direction is equal to half the sum of the absolute values ​​of the projections of the three sides onto the vertical direction. It should be noted that this calculation uses the sum of the absolute values ​​of the projections of each side rather than a precise OBB (directed bounding box) intersection. Although this may slightly overestimate the envelope height under extreme attitudes, the computational cost is extremely low and sufficient to meet the accuracy requirements of control training.

[0051] After obtaining half the height Afterwards, the lower boundary height of the connecting rod is ,in This represents the vertical component of the link's world coordinate position. If the current water level is... The original immersion ratio Calculate using the following formula:

[0052] in, This means cropping the values ​​to the range [0, 1]. To prevent small values ​​from being divided by zero (e.g., 1×10⁻⁶), -6 When the water level is much lower than the lower boundary of the connecting rod, This indicates that the connecting rod is completely in the air; when the water level is much higher than the upper boundary of the connecting rod... This indicates that the connecting rod is completely submerged; when the water level is between the upper and lower boundaries of the connecting rod... Values ​​are taken in a linear proportion.

[0053] To ensure continuous hydrodynamic changes during water inflow and outflow, the original submersion ratio was subjected to dead zone and exponential smoothing to obtain a smoothed submersion ratio. .

[0054]

[0055] in, This is the submersion dead zone parameter, used to eliminate the slight submersion ratio caused by numerical fluctuations when the connecting rod just contacts the water surface; The inlet and outlet water smoothness index is used to control the nonlinear growth curve of the submersion ratio.

[0056] In some embodiments of this disclosure, , .when The immersion ratio increases slowly in the early stage of immersion and increases rapidly when it is close to full immersion, which is consistent with the nonlinear change trend of the drainage volume of an actual object with the immersion depth.

[0057] For underwater vehicles or underwater robotic arms that are continuously submerged, if they are within the water body boundary, they can be directly ordered to... Therefore, the aforementioned envelope calculation and smoothing processing are unnecessary, reducing unnecessary computational overhead; if it leaves the water body region, it will be determined according to the water body boundary. Reduce to 0 or smoothly transition according to the boundary distance. For links that are completely above the water surface, It does not generate any hydrodynamic effect.

[0058] It should be noted that the above method for calculating the immersion ratio is based on the vertical envelope of an equivalent cuboid and is applicable to the case where an equivalent cuboid is selected in Example 2. If an equivalent ellipsoid or cylinder is selected as the equivalent geometric body, it can be replaced by the formula for the vertical envelope height of the corresponding geometric body. The calculation method remains the same, except for the dead zone and smoothing steps. Those skilled in the art can select the appropriate envelope calculation method according to the type of equivalent geometry.

[0059] For operation S140, the composite hydrodynamic force includes at least buoyancy and attitude-related drag. Buoyancy is the upward force acting on an object in a liquid, equal in magnitude to the weight of the liquid displaced by the object, and directed vertically upwards. Drag is the force acting on an object relative to the liquid, acting in the opposite direction of the relative motion, and its magnitude is proportional to the square of the relative velocity and the area facing the flow. For multi-link wading robots, the buoyancy and drag of each link differ due to variations in their immersion ratio, attitude, and relative velocity, requiring independent calculation for each link before composite analysis.

[0060] In some embodiments of this disclosure, the buoyancy direction is upward along the vertical direction of the world, for the first i Buoyancy of each link Calculate using the following formula:

[0061] in, For the first i The buoyancy scaling factor of each link is used to adjust the magnitude of buoyancy to adapt to different robot structures and training stability requirements. The density of water can be 998.2 kg / m³ (20°C clean water) in a preferred embodiment, but other values ​​can be used depending on the training scenario (such as seawater density of 1025 kg / m³). It is the acceleration due to gravity; For the first i The equivalent volume of each link is calculated from the equivalent side length in Example 2; To ensure a smooth immersion ratio; This is the unit vector pointing vertically upwards in the world. The physical meaning of this formula is that the buoyancy force on the connecting rod is equal to the product of water density, gravitational acceleration, and displacement volume (equivalent volume multiplied by the submersion ratio), and then multiplied by the buoyancy scaling factor for adjustment.

[0062] In some embodiments of this disclosure, for the calculation of attitude-related projected area drag, it is first necessary to construct the water flow velocity. The water flow can be constant (such as a uniform flow) or oscillating flow that varies with time or space (such as the movement of water particles caused by waves). The relative velocity of the connecting rod to the water flow is:

[0063] To avoid spikes in velocity-squared drag during transient velocity anomalies (e.g., when a connecting rod experiences collision rebound or numerical integral divergence), the following can be done: Each component or modulus is limited. Let the direction of the unit relative velocity be... The local coordinate axes of the connecting rod are , , Then the cosines of the included angles in the three directions are:

[0064] The projected areas of the equivalent cuboid along the normal directions of the three local coordinate axes are as follows:

[0065] in Let be the area of ​​the plane perpendicular to the x-axis. Let be the area of ​​the plane perpendicular to the y-axis. This is the area of ​​the surface perpendicular to the z-axis.

[0066] Attitude-related effective frontal area Calculate using the following formula:

[0067] The physical meaning of this formula is that the effective area of ​​the connecting rod facing the flow in the direction of relative velocity is equal to the weighted sum of the areas of the three surfaces, multiplied by the cosine of their angles with the incoming flow direction, to account only for the submerged portion. When one surface of the connecting rod faces the incoming flow, the cosine of that surface is close to 1, contributing the most; when one surface is parallel to the incoming flow direction, the cosine of the angle is close to 0, contributing the least.

[0068] Attitude-related drag Calculate using the following formula:

[0069] in For the first iThe drag coefficient of a connecting rod is related to the shape of the connecting rod and the Reynolds number, and can usually be taken as 0.8 to 1.2. This is the resistance scaling factor, used to adjust the overall resistance level. This represents the upper limit of the drag on a single link, used to prevent excessive drag under extreme conditions from causing numerical instability; the negative sign indicates that the direction of the drag is opposite to the direction of the relative velocity. It should be noted that the above calculations of buoyancy and drag are based on an equivalent cuboid. If, in the aforementioned embodiments, the link is chosen to be equivalent to an ellipsoid or cylinder, then the projected area... , , The calculation method needs to be replaced with the formula for the projected area of ​​the corresponding geometric object, while the remaining calculation steps remain unchanged. Additionally, the buoyancy scaling factor... and drag scaling factor Randomization can be performed through domain randomization to improve the robustness of the training strategy to uncertainties in hydrodynamic parameters.

[0070] In some embodiments of this disclosure, for target links with paddling, swinging, or maneuvering functions, the synthetic hydrodynamics also includes anisotropic paddling resistance: a local axis of the link is selected as a reference axis, and the relative water flow velocity is decomposed into a tangential component along the reference axis and a normal component perpendicular to the reference axis. Tangential resistance and normal resistance are calculated based on tangential resistance parameters and normal resistance parameters, respectively. Different types of target links are configured with different resistance parameters and gains.

[0071] It is understandable that for paddle-footed amphibious robots, biomimetic fin-shaped robots, surface-controlled robots, or operational robots with underwater robotic arms, the target link in water is not only subject to the passive resistance described in Example 4, but also generates direction-dependent reaction forces through sweeping, swinging, deflecting, or manipulating actions. This reaction force is a crucial mechanism for propulsion, steering, and braking in wading robots, and its core characteristic lies in anisotropy—that is, the magnitude of the hydrodynamic force experienced by the link varies when it moves in different directions. This invention expresses this characteristic through an anisotropic manipulatory resistance model.

[0072] First, select a local axis of the target link. The reference axis for paddling is chosen based on the motion and structural characteristics of the link. For example, for leg links, an axis along the length of the link can be selected as the paddling axis; for fin or rudder links, an axis along the fin's span can be selected; and for blade links, an axis along the blade's radius can be selected. The paddling reference axis can be individually specified for each target link through a configuration file.

[0073] The relative velocity of the connecting rod to the water flow is decomposed into tangential and normal components:

[0074]

[0075] in This indicates the relative velocity of the connecting rod along the paddle axis. This indicates the relative velocity of the connecting rod perpendicular to the paddle axis. Tangential motion corresponds to the direction in which the connecting rod "paddles downstream," while normal motion corresponds to the direction in which the connecting rod "paddles upstream."

[0076] Then, the tangential drag and normal drag are calculated separately, and anisotropic control drag is synthesized. :

[0077] in The tangential drag coefficient, This is the normal drag coefficient. For tangential reference area, For the normal reference area, For the first i The paddle gain of each link.

[0078] Normal drag coefficient Greater than the tangential drag coefficient This anisotropy results in a greater hydrodynamic reaction when the connecting rod moves perpendicular to the paddle axis, and less resistance when it moves along the paddle axis. This anisotropic characteristic is the physical basis of water propulsion: when the connecting rod paddles at a greater normal velocity, it generates a greater reaction force to create propulsion; when the connecting rod retracts tangentially, the resistance is less, reducing energy loss.

[0079] Different types of target links can be set for different categories such as thigh, calf, foot, fin, propeller blade, rudder surface, thruster mounting link or robotic arm end link. , , , and This is to express the differences in the ability of different links to generate propulsion, steering, braking, or disturbance in water. For example, the foot-end link, due to its larger area, can be set with a larger [size / size]. and To generate greater stroke reaction force; the thigh linkage, due to its smaller range of motion, can be set to a smaller value. To reduce unnecessary resistance.

[0080] For operating the S150, buoyancy, attitude-related drag, and anisotropic maneuvering drag are all calculated in the world coordinate system, while the external force interface of the 3D physical simulation platform typically requires input of forces and moments in the local coordinate system of the linkage. Therefore, the composite hydrodynamic forces need to be transformed into coordinates before being applied. Furthermore, to ensure numerical stability, numerical cleanup and amplitude limiting of the hydrodynamic forces and moments are also necessary.

[0081] In some embodiments of this disclosure, external forces and external torques are applied to the corresponding rigid body links via an interface, including: numerically cleaning and limiting the combined hydrodynamic forces and stress torques of each link, and then converting them to the link's local coordinate system; setting the hydrodynamic action point of the main link as the link origin, default local centroid, or user-specified point, and setting the hydrodynamic action point of non-main links as a cached default local centroid or user-configured point; and setting upper limits and absolute upper limits related to the link's mass for the combined hydrodynamic forces and stress torques, respectively.

[0082] For example, in each physical step, the resultant hydrodynamic force of the i-th link is:

[0083] The simulation environment first focuses on synthetic hydrodynamics. Numerical cleanup of stress torques was performed, replacing NaN, positive infinity, and negative infinity with zero because hydrodynamic calculations may produce non-finite values ​​under extreme conditions (such as abnormal velocities caused by collision rebounds), which could lead to physical simulation crashes if applied directly. Forces and torques were trimmed based on link weight and preset absolute upper limits. Specifically, a relative upper limit related to link mass and an absolute upper limit were set, with the smaller of the two values ​​used as the final limit. This prevents excessive hydrodynamic forces from causing the robot to fly away and accommodates links of different masses and sizes. Regarding the setting of the hydrodynamic application point, for fuselage or main body links, the hydrodynamic application point can be set to the link origin, the default local centroid, or a user-specified point; for non-main body links such as limbs, fins, propellers, rudder surfaces, thruster mounting links, or robotic arm links, the hydrodynamic application point is set to the default local centroid cached before domain randomization in Example 2 or a user-configured point. This approach preserves the hydrodynamic lever arm effect (i.e., the additional torque generated when the hydrodynamic force does not pass through the center of mass) while avoiding the randomization of the center of mass during operation, which causes the point of application of external forces to change with each randomization, thus ensuring the consistency and stability of hydrodynamic application.

[0084] It should be noted that the hydrodynamic application method of this invention is implemented based on the external force and external torque interface of a general-purpose three-dimensional physical simulation platform, without relying on a proprietary hydrodynamic module of a specific platform. This makes the method of this invention highly portable across platforms and can be implemented on various simulation platforms that support external force / torque interfaces, such as IsaacSim, Gazebo, and MuJoCo.

[0085] Figure 3 The diagram illustrates a configurable water surface for a wading robot.

[0086] In some embodiments of this disclosure, such as Figure 3 The configurable water surface shown includes at least one of the following: a horizontal static water surface, a dynamic water surface that changes over time, a wave water surface with a spatial phase distribution, and a water flow model with a velocity field; the water wading training environment also includes at least one of the following: land terrain, underwater terrain, water-land transition slope, water surface boundary, or underwater operation boundary.

[0087] Figure 4 The flowchart illustrates a method for training a motion control algorithm for a wading robot according to an embodiment of the present disclosure.

[0088] like Figure 4 As shown, the method may include operations S410-S420.

[0089] In operating the S410, a water-wading training environment with synthetic hydrodynamics is constructed in a three-dimensional physical simulation platform based on hydrodynamic simulation methods.

[0090] In operating the S420, based on the wading training environment after applying synthetic hydrodynamics, a data-driven control algorithm training task including domain randomization configuration is constructed to train the motion control strategy of the wading robot.

[0091] This hydrodynamic simulation method is provided according to any embodiment of this disclosure. Please refer to the description above for details regarding this hydrodynamic simulation method.

[0092] In some embodiments of this disclosure, the control algorithm training task configuration includes an observation space, an action space, a reward function, a termination condition, and a domain randomization configuration. The observation space includes at least one of robot body state, joint state, actuator state, task instructions, and environmental state estimation. The action space includes at least one of joint control variables, thruster control variables, and maneuvering surface control variables. The motion control strategy does not rely on discrete motion mode labels for explicit switching and can operate continuously across land, water-land transition, surface, and underwater scenarios. The control algorithm can employ reinforcement learning algorithms such as PPO, SAC, and TD3, or offline reinforcement learning, imitation learning, or other data-driven control methods.

[0093] For amphibious quadruped robots, the policy network input can include body angular velocity, gravity direction projection, velocity command, joint position, joint velocity, and previous action; the policy output can be the position target increment of each driven joint. For unmanned surface vessels, the policy input can include heading, yaw rate, track error, current estimation, and thruster state; the policy output can be thruster thrust or rudder angle. For underwater vehicles or underwater robots, the policy input can include depth, attitude, velocity, target relative pose, manipulator joint state, and current estimation; the policy output can be thruster thrust, rudder angle, manipulator joint target, or a combination thereof.

[0094] In some embodiments of this disclosure, the reward function includes at least one of velocity tracking, yaw rate tracking, airframe uprightness, depth maintenance, attitude stabilization, trajectory tracking, target approach, task completion, and energy consumption constraints. Penalty terms include at least one of joint position limits, joint velocity limits, vertical velocity, roll / pitch rate, rate of change of motion, joint torque, joint acceleration, thruster amplitude limiting, joint power, left-right power imbalance, and airframe linear acceleration. Domain randomization may be enabled during training, including factors such as ground friction, water density, water flow velocity, wave amplitude, airframe mass, non-airframe link mass, center of mass position, actuator stiffness, actuator damping, thruster efficiency, initial pose, and observation noise.

[0095] During training, domain randomization can be enabled, including at least one randomized parameter from the following: ground friction, water density, water flow velocity, wave amplitude, fuselage mass, non-fuselage link mass, center of mass position, actuator stiffness, actuator damping, thruster efficiency, initial pose, and observation noise. Domain randomization helps improve the robustness of training strategies and the ability to transfer simulations to real-world applications.

[0096] For example, the simulation physics step size is 0.005s, the action decimation is 4, and the policy frequency is 50Hz. The PPO trainer has a rollout length of 24 per environment, and both the actor and critic networks use [512, 256, 128] hidden layers, with ELU activation function, a learning rate of 1.0×10^-3, a discount factor of 0.99, a GAE parameter of 0.95, and a clip parameter of 0.2.

[0097] Figure 5 The schematic diagram illustrates a structural block diagram of a hydrodynamic simulation device 500 for a wading robot provided according to an embodiment of the present disclosure.

[0098] like Figure 5 As shown, the simulation device 500 includes: a model acquisition unit 501, an equivalent geometry buffer unit 502, an immersion ratio calculation unit 503, a hydrodynamic calculation unit 504, and a hydrodynamic application unit 505.

[0099] The model acquisition unit 501 is used to acquire a robot model, wherein the robot model includes fixed parameters. The model acquisition unit 501, for example, performs... Figure 1 Operation of S110 in the middle.

[0100] The equivalent geometry buffer unit 502 is used to determine and buffer the equivalent geometric parameters and hydrodynamic action points of each link based on the fixed parameters of each link in the model. For example, the equivalent geometry buffer unit 502 performs... Figure 1 Operation of S120 in the middle.

[0101] The submersion ratio calculation unit 503 is used to calculate the smooth submersion ratio of each link relative to the water surface at each physical step, based on the dynamic parameters and equivalent geometric parameters of each link. The submersion ratio calculation unit 503, for example, performs... Figure 1 Operation of S130 in the middle.

[0102] Hydrodynamic calculation unit 504 is used to calculate the combined hydrodynamic force of each link based on the smooth immersion ratio. Hydrodynamic calculation unit 504, for example, performs... Figure 1 Operation of S140 in the middle.

[0103] The hydrodynamic application unit 505 is used to transform the composite hydrodynamic force from the world coordinate system to the local coordinate system of each link to form the local hydrodynamic force of the link; and to apply the local hydrodynamic force of the link to the corresponding rigid body link for hydrodynamic simulation. The hydrodynamic application unit 505, for example, performs... Figure 1 Operation of S150 in the middle.

[0104] Figure 6 The schematic diagram illustrates a structural block diagram of a water-wading robot motion control algorithm training device 600 provided according to an embodiment of the present disclosure.

[0105] like Figure 6 As shown, the training device 600 includes an environment construction unit 610 and a training task construction unit 620.

[0106] Environment building unit 610 is used to build a water-based training environment with synthetic hydrodynamics in a three-dimensional physical simulation platform based on hydrodynamic simulation methods.

[0107] The training task construction unit 620 is used to construct a data-driven control algorithm training task with domain randomization configuration based on the wading training environment after applying synthetic hydrodynamics, and to train the motion control strategy of the wading robot.

[0108] The hydrodynamic simulation method is the hydrodynamic simulation method adopted in any embodiment of this disclosure.

[0109] For example, the model acquisition unit 501, the equivalent geometry cache unit 502, the immersion ratio calculation unit 503, the hydrodynamic calculation unit 504, and the hydrodynamic application unit 505 can be hardware, software, firmware, or any feasible combination thereof. For example, the model acquisition unit 501, the equivalent geometry cache unit 502, the immersion ratio calculation unit 503, the hydrodynamic calculation unit 504, and the hydrodynamic application unit 505 can be dedicated or general-purpose circuits, chips, or devices, or they can be a combination of a processor and memory. The embodiments of this disclosure do not limit the specific implementation of the above units.

[0110] It should be noted that in the embodiments of this disclosure, each unit of the hydrodynamic simulation device 500 for the wading robot corresponds to each step of the aforementioned construction method. For the specific functions of the hydrodynamic simulation device 500 for the wading robot, please refer to the relevant description of the construction method, which will not be repeated here. Figure 5 The components and structure of the hydrodynamic simulation device 500 for the wading robot shown are merely exemplary and not limiting. The hydrodynamic simulation device 500 for the wading robot may also include other components and structures as needed.

[0111] For example, the environment construction unit 610 and the training task construction unit 620 can be hardware, software, firmware, or any feasible combination thereof. For example, the environment construction unit 610 and the training task construction unit 620 can be dedicated or general-purpose circuits, chips, or devices, or a combination of a processor and memory. The embodiments of this disclosure do not limit the specific implementation of the above-mentioned units.

[0112] It should be noted that in the embodiments of this disclosure, each unit of the water robot motion control algorithm training device 600 corresponds to each step of the aforementioned output prediction method. For the specific functions of the water robot motion control algorithm training device 600, please refer to the relevant description of the output prediction method, which will not be repeated here. Figure 6 The components and structures of the wading robot motion control algorithm training device 600 shown are merely exemplary and not limiting. The wading robot motion control algorithm training device 600 may also include other components and structures as needed.

[0113] At least one embodiment of this disclosure also provides an electronic device including a processor and a memory, the memory including one or more computer program instructions. The one or more computer program instructions are stored in the memory and configured to be executed by the processor, and include instructions for implementing the above-described construction method or output prediction method. This electronic device can effectively solve the problems of performance instability and insufficient accuracy caused by random initialization in Extreme Learning Machines by applying the stellar oscillation optimization algorithm to the weight and bias optimization of Extreme Learning Machines, significantly improving the accuracy and stability of the model.

[0114] Figure 7 This is a schematic block diagram of an electronic device provided for some embodiments of this disclosure. For example... Figure 7 As shown, the electronic device 1200 includes a processor 1210 and a memory 1220. The memory 1220 stores non-transitory computer-readable instructions (e.g., one or more computer program modules). The processor 1210 executes the non-transitory computer-readable instructions, which, when executed by the processor 1210, can perform one or more steps of the construction method or output prediction method described above. The memory 1220 and the processor 1210 can be interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0115] For example, processor 1210 may be a central processing unit (CPU), a graphics processing unit (GPU), or other form of processing unit with data processing and / or program execution capabilities. For example, the central processing unit (CPU) may be an x86 or ARM architecture. Processor 1210 may be a general-purpose processor or a special-purpose processor, capable of controlling other components in electronic device 1200 to perform desired functions.

[0116] For example, memory 1220 may include any combination of one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, erasable programmable read-only memory (EPROM), portable compact disc read-only memory (CD-ROM), USB memory, flash memory, etc. One or more computer program modules may be stored on the computer-readable storage medium, and processor 1210 may run one or more computer program modules to implement various functions of electronic device 1200. Various application programs and various data, as well as various data used and / or generated by the application programs, may also be stored in the computer-readable storage medium.

[0117] It should be noted that, in the embodiments of this disclosure, the specific functions and technical effects of the electronic device 1200 can be referred to the description of the construction method or output prediction method above, and will not be repeated here.

[0118] Figure 8 This is a schematic block diagram of another electronic device provided for some embodiments of the present disclosure. This electronic device 1300 is, for example, suitable for implementing the construction method or output prediction method provided in the embodiments of the present disclosure. It should be noted that... Figure 8 The illustrated electronic device 1300 is merely an example and does not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0119] like Figure 8 As shown, electronic device 1300 may include a processing device (e.g., a central processing unit, a graphics processor, etc.) 1310, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1320 or a program loaded from storage device 1380 into random access memory (RAM) 1330. RAM 1330 also stores various programs and data required for the operation of electronic device 1300. Processing device 1310, ROM 1320, and RAM 1330 are interconnected via bus 1340. Input / output (I / O) interface 1350 is also connected to bus 1340.

[0120] Typically, the following devices can be connected to I / O interface 1350: input devices 1360 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 1370 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1380 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1390. Communication device 1390 allows electronic device 1300 to communicate wirelessly or wiredly with other electronic devices to exchange data. Although Figure 8 An electronic device 1300 with various devices is shown, but it should be understood that it is not required to implement or have all of the devices shown, and the electronic device 1300 may alternatively implement or have more or fewer devices.

[0121] For example, according to embodiments of this disclosure, the above-described construction method or output prediction method can be implemented as a computer software program. For instance, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program including program code for performing the above-described construction method or output prediction method. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 1390, or installed from a storage device 1380, or installed from a ROM 1320. When the computer program is executed by the processing device 1310, the functions defined in the construction method or output prediction method provided by embodiments of this disclosure can be implemented.

[0122] At least one embodiment of this disclosure also provides a computer-readable storage medium for storing non-transitory computer-readable instructions that, when executed by a computer, can implement the aforementioned construction or prediction method. Using this computer-readable storage medium, by applying the stellar oscillation optimization algorithm to the weight and bias optimization of the Extreme Learning Machine (ELM), the performance instability and insufficient accuracy of the ELM caused by random initialization can be effectively solved, significantly improving the model's accuracy and stability.

[0123] Figure 9 This is a schematic diagram of a storage medium provided for some embodiments of this disclosure. For example... Figure 9 As shown, storage medium 1400 is used to store non-transitory computer-readable instructions 1410. For example, when non-transitory computer-readable instructions 1410 are executed by a computer, one or more steps in the construction or prediction methods described above can be performed.

[0124] For example, the storage medium 1400 can be used in the aforementioned electronic device 1200. For example, the storage medium 1400 can be... Figure 9 The memory 1220 in the illustrated electronic device 1200. For example, a description of the storage medium 1400 can be found here. Figure 9 The corresponding description of the memory 1220 in the illustrated electronic device 1200 will not be repeated here.

[0125] The above are merely specific embodiments of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A hydrodynamic simulation method for a wading robot, comprising: Obtain the robot model, which includes fixed parameters; Based on the fixed parameters of each link in the model, determine and buffer the equivalent geometric parameters and hydrodynamic action points of each link; In each physical step, the smooth submersion ratio of each link relative to the water surface is calculated based on the dynamic parameters of each link and the equivalent geometric parameters. Based on the smooth immersion ratio, the combined hydrodynamic force of each link is calculated; The composite hydrodynamic force is transformed from the world coordinate system to the local coordinate system of each link to form the local hydrodynamic force of the link; the local hydrodynamic force of the link is applied to the corresponding rigid body link to perform hydrodynamic simulation.

2. The construction method according to claim 1, wherein, Determining the equivalent geometric parameters of each link includes: Each of the connecting rods is equivalent to a regular geometric body with a uniform mass distribution; The equivalent geometric dimensions are inversely derived based on the default mass of each link and the analytical relationship between the inertia of the regular geometry, and the equivalent volume of each link is calculated.

3. The construction method according to claim 2, wherein, The calculation of the combined hydrodynamic force of each link includes: The envelope of the link in the vertical direction is calculated based on the link attitude and equivalent geometric parameters. The original submersion ratio is obtained based on the relative relationship between the water surface position and the link envelope. Dead zone processing and nonlinear smoothing are applied to the original immersion ratio to obtain a smoothed immersion ratio; for the connecting rod that is continuously within the water body boundary, the immersion ratio is set to full immersion; for the connecting rod that is completely above the water surface, the immersion ratio is set to zero. The synthetic hydrodynamics include at least buoyancy and attitude-related drag.

4. The method according to claim 3, wherein, The buoyancy is calculated based on the buoyancy scaling factor, water density, gravitational acceleration, equivalent volume of the connecting rod, and smooth immersion ratio, with the buoyancy direction being upward along the vertical direction of the world.

5. The method according to claim 3, wherein, The attitude-related drag includes: Calculate the relative velocity of the connecting rod with respect to the water flow, determine the effective projected area of ​​the connecting rod in the direction of relative velocity based on the connecting rod's attitude, and calculate the resistance according to the velocity square resistance model; set upper limits for the modulus of relative velocity and the amplitude of single connecting rod resistance respectively.

6. The method according to claim 5, wherein, For target linkages with paddling, swinging, or maneuvering functions, the composite hydrodynamics also include anisotropic paddling drag: A local axis of the link is selected as the reference axis, and the relative water flow velocity is decomposed into a tangential component along the reference axis and a normal component perpendicular to the reference axis. Tangential resistance and normal resistance are calculated based on tangential resistance parameters and normal resistance parameters, respectively. Different types of target links are configured with different resistance parameters and gains.

7. The method according to claim 1, wherein, The process of applying local hydrodynamic forces to the corresponding rigid body link includes: After numerical cleanup and amplitude limiting of the combined hydrodynamic forces and stress moments of each link, the data are transformed to the local coordinate system of the link. Set the hydrodynamic application point of the main link to the link origin, the default local centroid, or the user-specified point, and set the hydrodynamic application point of the non-main link to the cached default local centroid or the user-configured point. Upper limits and absolute limits related to the connecting rod mass are set for the combined hydrodynamic force and the stress torque, respectively.

8. A method for training a motion control algorithm for a wading robot, comprising: Based on hydrodynamic simulation methods, a water-wading training environment with synthetic hydrodynamics is constructed in a three-dimensional physical simulation platform; Based on the wading training environment after applying the synthetic hydrodynamics, a data-driven control algorithm training task including domain randomization configuration is constructed to train the motion control strategy of the wading robot. The hydrodynamic simulation method described herein is the hydrodynamic simulation method according to any one of claims 1-7.

9. An electronic device, comprising: processor; A memory, comprising one or more computer program instructions, wherein the one or more computer program instructions are stored in the memory and, when executed by the processor, implement the method of any one of claims 1-8.

10. A computer-readable storage medium that non-transitoryly stores computer-readable instructions, wherein, The method of any one of claims 1-8 is implemented when the computer-readable instructions are executed by a processor.