Underwater robot control method and device and underwater robot control system

Abstract

This invention provides an underwater robot control method, device, and control system, belonging to the field of underwater intelligent operation technology. The method includes: acquiring multimodal perception data generated by an operator operating a remotely controlled device; performing data prediction on the multimodal perception data to determine the operator's motion intention and predicted reference trajectory; acquiring the link geometry parameters of the robotic arm; performing hydrodynamic calculations on the link geometry parameters to obtain a fluid feedforward compensation torque; acquiring the rigid body foundation control torque of the robotic arm; superimposing the rigid body foundation control torque and the fluid feedforward compensation torque to obtain a total driving torque; and driving the robotic arm to perform trajectory tracking based on the total driving torque to execute underwater operations. The above solution can effectively improve the trajectory tracking accuracy of the underwater robotic arm and the execution effect of underwater operations.

Description

Technical Field

[0001] This invention relates to the field of underwater intelligent operation technology, specifically to underwater robot control methods, devices, and control systems. Background Technology

[0002] Underwater robots are core equipment for marine exploration and resource development, playing an irreplaceable role in deep-sea operations, oil and gas engineering, environmental monitoring, and emergency rescue. The remote control accuracy of the underwater robotic arm directly determines the success rate of underwater missions. Existing technologies typically control the robotic arm directly by acquiring sensor data and calculate the control torque by combining buoyancy and gravity in equivalent processing during dynamic modeling.

[0003] However, in practical applications, existing technologies lack data prediction based on perceived data, making it difficult to accurately determine the operator's motion intentions and predict reference trajectories. Furthermore, they cannot distinguish between fluid disturbances and do not independently feedforward to compensate for viscous drag torque, additional mass torque, and separately solved buoyancy torque, resulting in the robotic arm drifting due to fluid disturbances when performing trajectory tracking.

[0004] Therefore, there is an urgent need for a control method for underwater robotic arms that can perform data prediction and independent feedforward compensation based on perceived data. Summary of the Invention

[0005] To address the aforementioned problems, this application provides an underwater robot control method to improve the trajectory tracking accuracy of underwater robotic arms and the effectiveness of underwater operations. The technical solution is as follows: In a first aspect, the present invention provides an underwater robot control method, applied to an underwater robot equipped with a robotic arm for performing underwater operations. The underwater robot is connected to a remote control device via a bidirectional long-delay communication link. The remote control device is used by an operator to manually control the robotic arm, including: Acquire multimodal sensing data generated by the operator when operating the remote control device, perform data prediction on the multimodal sensing data, determine the operator's motion intention and predict reference trajectory; Obtain the link geometry parameters of the robotic arm, perform fluid dynamics calculations on the link geometry parameters, and obtain the fluid feedforward compensation torque. The fluid feedforward compensation torque is the total fluid dynamic load including viscous drag torque, additional mass torque, and separately solved buoyancy torque. Obtain the rigid body foundation control torque of the robotic arm, and superimpose the rigid body foundation control torque with the fluid feedforward compensation torque to obtain the total driving torque; The total driving torque drives the robotic arm to perform trajectory tracking in order to carry out underwater operations.

[0006] Combining the first aspect and the above implementation methods, in some possible implementation methods, the fluid dynamics parameters of the connecting rod are calculated to obtain the fluid feedforward compensation torque, including: The homogeneous transformation matrix of the link's geometric parameters is solved to obtain the link's centroid Jacobian matrix. Based on the Morrison equation, the Jacobian matrix of the connecting rod's center of mass is integrated to obtain the viscous drag torque and the additional mass torque. The buoyancy of a single link is mapped based on the Jacobian matrix of the link's center of mass to obtain the buoyancy torque that can be solved separately. The fluid feedforward compensation torque is obtained by superimposing the viscous drag torque, the additional mass torque, and the separately solved buoyancy torque.

[0007] Combining the first aspect and the above implementation methods, in some possible implementation methods, data prediction is performed on multimodal sensing data to determine the operator's motion intention and predict reference trajectory, including: Multimodal sensing data is subjected to composite filtering to obtain noise-reduced sensing data; Spatiotemporal registration and fusion processing is performed based on noise-reduced sensing data to obtain unified spatiotemporal aligned fusion features; Intent-time reasoning and trajectory prediction are performed on the fusion features of unified spatiotemporal alignment to determine the operator's motion intent and predict the reference trajectory. The predicted reference trajectory is a future trajectory used to compensate for the latency of bidirectional long-delay communication links.

[0008] Combining the first aspect and the above implementation methods, in some possible implementation methods, before performing composite filtering on the multimodal sensing data to obtain the denoised sensing data, the following steps are included: Acquire the amplitude of the operator's electromyographic signal; Fatigue state values ​​are obtained by quantifying the amplitude of electromyographic signals. When the fatigue state value exceeds the preset threshold, a smooth switching process for control weights is performed to obtain dynamic human-machine hybrid control weight instructions. These instructions are used to adjust the allocation ratio between manual control by the operator and autonomous control by the underwater robot.

[0009] Combining the first aspect and the above implementation methods, in some possible implementation methods, after driving the robotic arm to perform trajectory tracking based on the total driving torque to execute a preset underwater operation, the following steps are included: Acquire underwater interaction forces collected by the end effector force sensor of the robotic arm; The underwater interactive force is transmitted back to the remote control equipment via a two-way long-delay communication link to provide force feedback prompts to the operator.

[0010] In combination with the first aspect and the above implementation methods, some possible implementation methods also include: Acquire transmission data from a two-way long-latency communication link. The transmission data includes the predicted reference trajectory and the joint states and end effector interaction forces of the robotic arm. Based on the wave variable energy shaping model, passive constraint processing is performed on the transmitted data to obtain wave variable signals that satisfy energy conservation constraints. The wave variable signals are used to suppress system oscillations caused by communication delay. The wave variable signal is transmitted bidirectionally through a bidirectional long-delay communication link.

[0011] In combination with the first aspect and the above implementation methods, some possible implementation methods also include: Acquire samples of ocean current disturbances at multiple speeds; The data generated from underwater operations, the predicted reference trajectory, and the samples of multi-velocity ocean current disturbances are superimposed and amplified to obtain the amplified training samples. The dynamic motion primitive model is updated and corrected based on the expanded training samples to generate skill trajectories. The dynamic motion primitive model is used to generate stable operation trajectories that are resistant to ocean current disturbances. The skill trajectories are stored in layers to obtain a trajectory skill library.

[0012] Combining the first aspect and the above implementation methods, in some possible implementations, the skill trajectories are stored in layers to obtain a trajectory skill library, which includes: Obtain the mission trajectory of the underwater mission to be processed; The task trajectory is searched and matched with the trajectory skill library to obtain the matching skill trajectory; The matching skill trajectory is adaptively corrected based on a real-time fluid dynamics model to generate a target execution trajectory, which is then used to control the robotic arm to perform the underwater task to be processed.

[0013] Secondly, the present invention also provides an underwater robot control device, characterized in that it is applied to an underwater robot, the underwater robot being equipped with a robotic arm for performing underwater operations, the underwater robot being connected to a remote control device via a bidirectional long-delay communication link, the remote control device being used by an operator to manually control the robotic arm, including: The data prediction unit is used to acquire multimodal sensing data generated by the operator when operating the remote control device, perform data prediction on the multimodal sensing data, determine the operator's motion intention and predict the reference trajectory; The compensation torque acquisition unit is used to acquire the link geometry parameters of the robotic arm, perform fluid dynamics calculations on the link geometry parameters, and obtain the fluid feedforward compensation torque. The fluid feedforward compensation torque is the total fluid dynamic load including viscous drag torque, additional mass torque, and separately solved buoyancy torque. The total driving torque acquisition unit is used to acquire the rigid body base control torque of the robotic arm, and to superimpose the rigid body base control torque with the fluid feedforward compensation torque to obtain the total driving torque. The work execution unit is used to drive the robotic arm to perform trajectory tracking based on the total driving torque in order to perform underwater operations.

[0014] Thirdly, the present invention also provides an underwater robot control system, including a communication module, a sensing component, a robotic arm drive component, a memory, and a processor, wherein... The communication module is used to receive multimodal perception data from the operator via a two-way long-latency communication link, and to send the joint status and end effector interaction force of the robotic arm. Sensing components are used to acquire the link geometry parameters and end effector forces of the robotic arm; A robotic arm drive assembly is used to drive the robotic arm to perform trajectory tracking for underwater operations under the control of a processor; Memory, used to store executable program code; The processor is used to perform data prediction on multimodal sensing data, determine the operator's motion intention and predict the reference trajectory. The predicted reference trajectory is a future trajectory used to compensate for the latency of the two-way long-delay communication link. It performs hydrodynamic calculations on the linkage geometry parameters to obtain the fluid feedforward compensation torque. It acquires the rigid body base control torque of the robotic arm, and superimposes the rigid body base control torque with the fluid feedforward compensation torque to obtain the total driving torque. Based on the total driving torque, it controls the robotic arm drive assembly to drive the robotic arm to perform trajectory tracking to execute preset underwater operations.

[0015] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps of the underwater robot control method in any of the above implementations.

[0016] The beneficial effects of this invention are as follows: By collecting multimodal sensing data generated during the operation of remote control equipment by the operator, and performing time-series prediction calculations based on the multimodal sensing data, the operator's motion intention is accurately identified and a predicted reference trajectory is output, avoiding the shortcomings of insufficient motion intention recognition accuracy in traditional solutions. Addressing the error drawbacks caused by the equivalent calculation of buoyancy and gravity in existing technologies, this application collects the geometric parameters of each link of the underwater robotic arm, and completes the calculation based on a fluid dynamics model to obtain the fluid feedforward compensation torque. This compensation torque is composed of viscous drag torque, additional mass torque, and independently calculated buoyancy torque, comprehensively characterizing the total fluid dynamic load of the underwater environment. Subsequently, the rigid body foundation control torque of the robotic arm is obtained, and the rigid body foundation control torque is superimposed with the fluid feedforward compensation torque to synthesize the total driving torque. Finally, the total driving torque drives the underwater robotic arm to complete trajectory tracking and underwater operation tasks, significantly suppressing trajectory drift caused by fluid disturbances and significantly improving the trajectory tracking accuracy and underwater operation performance of the underwater robotic arm. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic diagram illustrating the application framework of the underwater robot control method provided by the present invention; Figure 2 This is a flowchart illustrating the underwater robot control method provided by the present invention. Figure 3 For the present invention Figure 2 A schematic diagram of an embodiment of S202; Figure 4 For the present invention Figure 2 A schematic diagram of an embodiment of S201; Figure 5 For the present invention Figure 2 A schematic diagram of an embodiment of S204; Figure 6 For the present invention Figure 2 A schematic diagram of an embodiment of S204; Figure 7 A schematic diagram illustrating the construction process of the trajectory skill library provided by this invention; Figure 8 A schematic diagram of the architecture of the Long Short-Term Memory network provided by this invention; Figure 9 This is a schematic diagram of the underwater robot control device provided by the present invention; Figure 10 This is a schematic diagram of the underwater robot control system provided by the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0021] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] This application provides an underwater robot control method, wherein the execution subject of the underwater robot control method is an underwater robot control method device or an underwater robot control system equipped with an underwater robot control method device. The following provides a detailed description; it should be noted that the order of description of the following embodiments is not intended to limit the preferred order of the embodiments.

[0024] Please refer to the following: Figure 1 , Figure 1 This is a schematic diagram illustrating the application framework of the underwater robot control method provided by this invention. For example... Figure 1 As shown, there are four hardware and software modules: remote control equipment, communication transmission unit, underwater robot, and virtual-real simulation skills training platform. The modules interact jointly through real-time control signals and offline iterative data. The specific collaborative process is as follows: The remote-controlled device serves as the front end for human-machine interaction, allowing the operator to input control commands. An electromyography (EMG) acquisition module collects the operator's surface EMG signals, while a 6-axis feedback manipulator collects hand-operated manipulation forces. The acquired multimodal sensing data is input into a composite filtering unit for noise reduction, and then undergoes multimodal fusion processing to obtain unified features. These fused features are then fed into a trajectory prediction unit to generate a predictive reference trajectory, and into a fatigue detection module to assess the operator's fatigue state for dynamic permission allocation. A human-machine visualization display shows the feedback torque, underwater images, and force feedback.

[0025] The communication transmission unit is responsible for overcoming the long latency of underwater acoustic signals to achieve bidirectional data transmission. It transmits downlink commands, such as the predicted reference trajectory generated by the remote control equipment, to the underwater robot, while simultaneously transmitting uplink feedback data from the underwater robot back to the remote control equipment.

[0026] The underwater robot serves as the core of the system, performing real-time control. The trajectory planning layer receives the predicted reference trajectory for descent and generates a continuous, smooth joint trajectory that satisfies velocity and acceleration constraints. The fluid dynamics modeling and calculation layer, combined with real-time seawater parameters acquired by the seawater parameter acquisition module, independently calculates the fluid feedforward compensation torque. The feedforward compensation torque controller superimposes the rigid body foundation control torque and the fluid feedforward compensation torque to obtain the total driving torque. The execution layer drives the robotic arm to perform underwater operations based on the total driving torque. A six-axis force sensor collects underwater interaction forces in real time and uploads them to the communication transmission unit via a data feedback module.

[0027] The virtual-real simulation skills training platform jointly optimizes underwater robots using offline iterative data. It overlays and expands real operational data with simulated multi-velocity ocean current disturbance samples to update and correct the dynamic motion primitive model, generating skill trajectories resistant to ocean current disturbances and storing them in layers. This assists underwater robots in achieving cross-task skill transfer and non-repetitive teaching operations.

[0028] The fourth layer is a virtual-real simulation skills training platform, which performs joint optimization with the robot's control unit through offline training / iterative data. The Computational Fluid Dynamics (CFD) flow field disturbance simulation module generates multi-velocity disturbance samples, and combines them with real operation data input to improve the Dynamic Movement Primitive (DMP) trajectory generation module to perform realistic fluid disturbance modeling and dynamic movement primitive optimization, generating disturbance-resistant skill trajectories. The Inverse Reinforcement Learning (IRL) policy transfer alignment module performs inverse reinforcement virtual-real correction to eliminate simulation and reality deviations. Finally, the skill trajectories are stored in a hierarchical skill library containing primitive / composite / task-level levels, and the skill library parameters are sent to the underwater robot's control unit to assist in subsequent trajectory generation and execution.

[0029] Please see Figure 2 , Figure 2 This is a flowchart illustrating the underwater robot control method provided in an embodiment of this application. Figure 2 As shown, the method in this application embodiment may include the following steps S201-S204: S201: Acquire multimodal sensing data generated by the operator when operating the remote control device, perform data prediction on the multimodal sensing data, and determine the operator's motion intention and predict the reference trajectory.

[0030] In this embodiment, multimodal sensing data refers to a collection of heterogeneous sensor signals generated by the operator when operating the remote-controlled device, including surface electromyography signals, force feedback manipulation data, and master hand joint motion data. The remote-controlled device refers to a hardware device for the operator to remotely input control commands, specifically including a 6-axis force feedback manipulator and a master hand joint encoder. Motion intention refers to the desired motion state and operational trend that the operator expects the underwater robotic arm to achieve. The predicted reference trajectory refers to a joint motion reference trajectory for a future period of time, predicted based on historical operational data to compensate for underwater acoustic communication delays.

[0031] In one feasible implementation, the multimodal sensing and acquisition subunit in the remote control device collects surface electromyography signals of the operator's upper limbs, hand interaction manipulation forces, and the trajectory of the main hand's joints in real time, forming multimodal sensing data. Subsequently, the multimodal sensing data undergoes data prediction processing, specifically including noise reduction and fusion of the multimodal sensing data, extracting features reflecting the operator's true control intentions, and performing temporal reasoning and trajectory prediction based on these features to determine the operator's movement intentions and the predicted reference trajectory for the next 3 seconds, so as to issue commands in advance to compensate for the long latency of underwater acoustic communication.

[0032] S202: Obtain the link geometry parameters of the robotic arm, perform fluid dynamics calculations on the link geometry parameters, and obtain the fluid feedforward compensation torque.

[0033] In this embodiment, the link geometry parameters refer to the geometric relationship parameters between adjacent links of a six-degree-of-freedom underwater manipulator described using the standard Denavit-Hartenberg four-parameter method. Hydrodynamic solution refers to the real-time calculation process of independently solving for the torques generated by viscous drag, additional mass force, and buoyancy based on the link geometry parameters and seawater environment parameters. The fluid feedforward compensation torque refers to the total hydrodynamic load composed of the superposition of the viscous drag torque, the additional mass torque, and the separately solved buoyancy torque, used for feedforward compensation at the control layer.

[0034] In one feasible implementation, the link geometry parameters of the robotic arm are obtained through the fluid dynamics modeling and calculation layer of the underwater robot's control end. The homogeneous transformation matrix is ​​solved step by step using the standard Denavit-Hartenberg four-parameter method to obtain the Jacobian matrix of the centroid linear velocity of each link.

[0035] The viscous drag torque and additional mass torque are solved by integrating the Morrison equation along the connecting rod axis. Simultaneously, the buoyancy of each connecting rod is calculated independently and mapped to the joint buoyancy torque using the Jacobian transpose matrix. The viscous drag torque, additional mass torque, and buoyancy torque are then superimposed to obtain the fluid feedforward compensation torque. S203: Obtain the rigid body foundation control torque of the robotic arm, and superimpose the rigid body foundation control torque with the fluid feedforward compensation torque to obtain the total driving torque.

[0036] In the embodiments of this application, the rigid body foundation control torque refers to the control torque output by the calculated torque controller based on the rigid body dynamics model under conditions of no fluid disturbance.

[0037] In one feasible implementation, the feedforward compensation calculation torque control layer obtains the rigid body foundation control torque calculated based on the Lagrange rigid body dynamics equation at each discrete time k, and superimposes the rigid body foundation control torque with the fluid feedforward compensation torque solved in real time.

[0038] For example, the specific form of the superposition processing formula in step S203 above is as follows:

[0039] Among them, formula parameters Representing the Total driving torque at discrete moments, formula parameters Representing the The rigid body foundation control torque at discrete moments, formula parameters Representing the The fluid feedforward compensation torque at discrete moments is obtained by superimposing the above values ​​to obtain the total driving torque, thereby compensating for fluid disturbances in real time based on rigid body control.

[0040] S204 drives the robotic arm to perform trajectory tracking based on the total driving torque to execute underwater operations.

[0041] In this embodiment of the application, trajectory tracking refers to the process by which the servo drivers of each joint of the robotic arm drive the joints to move according to the total driving torque command, so that the pose of the end effector approximates the predicted reference trajectory. In one feasible implementation, the total driving torque command is sent to the servo drives of each joint of the six-DOF underwater robotic arm via servo actuators, driving the robotic arm to perform trajectory tracking to complete the underwater operation. Simultaneously, a six-axis force sensor at the end effector collects underwater interaction forces in real time and transmits them back to the remote control equipment via an underwater acoustic communication transmission unit, thus forming a dynamic control closed loop.

[0042] In summary, this application collects multimodal sensing data generated during the operation of remote control equipment by the operator, performs time-series prediction calculations based on the multimodal sensing data, accurately identifies the operator's motion intentions, and outputs a predicted reference trajectory, avoiding the shortcomings of insufficient motion intention recognition accuracy in traditional solutions. Addressing the errors caused by the combined equivalent calculation of buoyancy and gravity in existing technologies, this application collects the geometric parameters of each link of the underwater robotic arm, solves the problem using a fluid dynamics model, and obtains the fluid feedforward compensation torque. This compensation torque is composed of viscous drag torque, additional mass torque, and independently calculated buoyancy torque, comprehensively characterizing the total fluid dynamic load of the underwater environment. Subsequently, the rigid body foundation control torque of the robotic arm is obtained, and the rigid body foundation control torque is superimposed with the fluid feedforward compensation torque to synthesize the total driving torque. Finally, the total driving torque drives the underwater robotic arm to complete trajectory tracking and underwater operation tasks, significantly suppressing trajectory drift caused by fluid disturbances and significantly improving the trajectory tracking accuracy and underwater operation performance of the underwater robotic arm.

[0043] Please see Figure 3 , Figure 3 This is the present invention. Figure 2 A schematic diagram of an embodiment of S202. (See attached diagram.) Figure 3 As shown, the method in this application embodiment may include the following steps S301-S304: S301, the homogeneous transformation matrix of the connecting rod geometric parameters is solved to obtain the Jacobian matrix of the connecting rod's centroid.

[0044] In this embodiment, the homogeneous transformation matrix refers to a 4x4 matrix constructed based on the standard Denavit-Hartenberg four-parameter method to describe the coordinate transformation relationship between adjacent links. The link center-of-mass Jacobian matrix refers to the Jacobian matrix that maps joint velocities to the linear velocities of the link center-of-mass.

[0045] In one feasible implementation, the linkage geometry is described using the standard Denavit-Hartenberg four-parameter method via the DH kinematics submodule. The end-point base pose is solved by successive multiplication of homogeneous transformation matrices, deriving analytical solutions for both forward and inverse kinematics. Based on this, the forward kinematic equations are differentiated, and the Jacobian matrix of the center-of-mass velocity of each linkage is output synchronously, providing a foundation for subsequent fluid torque mapping calculations.

[0046] S302, based on the Morrison equation, the Jacobian matrix of the connecting rod's center of mass is integrated to obtain the viscous drag torque and the additional mass torque.

[0047] Morrison's equations are hydrodynamic equations used to calculate the viscous drag and added mass force experienced by slender underwater cylinders under the influence of waves and flow fields. Viscous drag torque refers to the torque exerted on the joint by the resistance caused by viscous friction as the fluid flows over the surface of a robotic arm link. Added mass torque refers to the torque exerted on the joint by the inertial force generated when the link accelerates in the fluid, causing the surrounding fluid to accelerate as well.

[0048] In one feasible implementation, the fluid torque solver module solves for the viscous drag torque and the additional mass torque by integrating the Morrison equation along the connecting rod axis.

[0049] S303 maps the buoyancy of a single link based on the Jacobian matrix of the link's center of mass, and obtains the buoyancy torque that can be solved separately.

[0050] In this embodiment, the buoyancy of a single link refers to the upward buoyancy force on a single link in water, calculated according to Archimedes' principle. The buoyancy torque refers to the torque generated when the buoyancy of a single link is mapped to the joint space through the Jacobian transpose matrix of the link's center of mass. In one feasible implementation, buoyancy is modeled separately by a fluid torque solver submodule, rather than being combined with gravity for equivalent processing.

[0051] For example, the specific form of the single-link buoyancy formula in step S303 above is as follows:

[0052] Among them, formula parameters Representing the Buoyancy of a single link in a series of links, formula parameters Represents seawater density, formula parameters Represents gravitational acceleration, formula parameters Representing the The displacement volume of each link is then mapped to the joint buoyancy torque via the Jacobian transpose matrix of the link's center of mass. For example, the mapping formula takes the following form:

[0053] Among them, formula parameters Representing the The joint buoyancy torque of each link, formula parameters Representing the The transpose of the Jacobian matrix of the linear velocity of the center of mass of each link, formula parameters Representing the The buoyancy of each link is calculated individually. The total buoyancy torque is obtained by summing the buoyancy torque of all links, which is the buoyancy torque that can be solved individually.

[0054] S304, the viscous drag torque, the additional mass torque, and the separately solved buoyancy torque are superimposed to obtain the fluid feedforward compensation torque. In one feasible implementation, the viscous drag torque obtained by integral solution based on the Morrison equation, the additional mass torque, and the separately solved and mapped buoyancy torque are superimposed to obtain the fluid feedforward compensation torque.

[0055] For example, the specific form of the superposition formula in step S304 above is as follows:

[0056] Among them, formula parameters Represents the fluid feedforward compensation torque, formula parameters Represents the viscous resistance torque, formula parameters Represents the additional mass torque, formula parameters This represents the buoyancy torque solved separately. In summary, this application achieves homogeneous variation of the link's geometric parameters.

[0057] In summary, this application achieves independent separation and refined solution of buoyancy, viscous drag, and additional mass force by solving the link geometry parameters through homogeneous transformation and Jacobian mapping, and by solving the viscous drag, additional mass force, and buoyancy torque separately based on Morrison's equations and independent buoyancy modeling, and then superimposing them. This avoids the wrist joint disturbance distortion caused by the equivalence of buoyancy and gravity, and improves the accuracy of fluid torque feedforward compensation.

[0058] In one feasible implementation, before performing composite filtering on the multimodal sensing data to obtain denoised sensing data, the following is also specifically performed: In the embodiments of this application, the electromyographic signal amplitude refers to the root mean square amplitude of the surface electromyographic signal, which is used to characterize the degree of muscle activation of the operator.

[0059] In one feasible implementation, the dynamic permission decision-making subunit acquires the operator's surface electromyography (EMG) signal in real time and calculates its root mean square amplitude as the EMG signal amplitude.

[0060] Fatigue state values ​​are obtained by quantifying the amplitude of electromyographic signals.

[0061] In this embodiment, fatigue quantification refers to the process of assessing the operator's muscle fatigue level based on the root mean square amplitude of electromyography (EMG) signals. The fatigue state value refers to a numerical value used to quantify the operator's fatigue level.

[0062] In one feasible implementation, the dynamic permission decision-making subunit calculates the root mean square amplitude of the electromyography signal in real time to quantify the operator's fatigue level and obtain the fatigue state value.

[0063] When the fatigue state value exceeds the preset threshold, a smooth switching process for control weights is performed to obtain dynamic human-machine hybrid control weight instructions. These instructions are used to adjust the allocation ratio between manual control by the operator and autonomous control by the underwater robot.

[0064] In this embodiment, the smooth switching of control weights refers to the process of adjusting the human-machine control weights based on Lyapunov stability constraints to ensure system stability when the operator's fatigue level exceeds a threshold. The dynamic human-machine hybrid control weight instruction refers to an instruction that includes the allocation ratio between the operator's manual control weights and the underwater robot's autonomous control weights. In one feasible implementation, a fatigue threshold greater than 0.5mV is set as a preset threshold. When the fatigue state value exceeds the preset threshold, the human-machine control weight is smoothly switched through Lyapunov stability constraints to generate dynamic human-machine hybrid control weight instructions.

[0065] For example, when the operator is fatigued, the robot's autonomous control weight is smoothly increased from 0.7 to 0.9, reducing the operator's manual control weight, thereby suppressing human-machine coupling oscillations. In summary, this application achieves adaptive dynamic allocation of human-machine control authority by quantifying fatigue state based on electromyographic signal amplitude and smoothly switching control weights based on Lyapunov constraints during fatigue. This avoids system instability caused by the decline in operator efficiency over long periods of operation, and improves the reliability of operation and work efficiency.

[0066] Please see Figure 4 , Figure 4 This is the present invention. Figure 2 A schematic diagram of an embodiment of S201. (See attached diagram.) Figure 4 As shown, the method in this application embodiment may include the following steps S401-S403: S401 performs composite filtering on the multimodal sensing data to obtain noise-reduced sensing data.

[0067] In this embodiment, composite filtering refers to the process of filtering out mixed noise in the sensing signal step by step using a series two-stage composite filtering strategy. Noise-reduced sensing data refers to sensing data with improved signal-to-noise ratio after composite filtering.

[0068] In one feasible implementation, the multimodal sensing acquisition subunit is equipped with a series two-stage composite filtering unit to perform composite filtering processing on the acquired multimodal sensing data. The first stage performs median filtering to remove pulse spikes and outliers caused by ocean current impacts. The second stage performs wavelet multi-scale decomposition and high-frequency threshold noise reduction on the filtered signal, filtering out mixed sensing noise introduced by seawater attenuation and ocean currents step by step. Overall, the signal-to-noise ratio of the electromyography signal can be improved by 20dB to obtain noise-reduced sensing data.

[0069] S402 performs spatiotemporal registration and fusion processing based on noise-reduced sensing data to obtain unified spatiotemporal aligned fusion features.

[0070] In this embodiment, spatiotemporal registration fusion processing refers to the fusion processing of heterogeneous multimodal sensor data by using cascaded time-varying Kalman-particle filtering for time synchronization and spatial alignment. The unified spatiotemporal alignment fusion feature refers to the unified multimodal feature that eliminates nonlinear non-Gaussian bias after spatiotemporal registration fusion.

[0071] In one feasible implementation, time-varying Kalman filtering is used to perform temporal state pre-estimation on three noise-reduced sensing data channels: electromyography, force feedback, and main hand joint trajectory, respectively, and outputs low-noise linear state variables. The state mean and covariance output by Kalman filtering are used as the initial distribution of the particle set. The observation nonlinearity and non-Gaussian bias caused by the underwater flow field are compensated by particle filtering resampling, thereby outputting unified spatiotemporal aligned fusion features.

[0072] S403 performs intent temporal reasoning and trajectory prediction on the fusion features of unified spatiotemporal alignment to determine the operator's motion intent and predict reference trajectory.

[0073] In this embodiment of the application, intent temporal reasoning refers to the process of inferring the operator's temporal motion intent through a dynamic Bayesian network based on the fusion features of unified spatiotemporal alignment.

[0074] In one feasible implementation, the unified spatiotemporally aligned fusion features are fed into a dynamic Bayesian network to achieve temporal reasoning of the operator's motion intention and determine the operator's motion intention. Simultaneously, the long short-term memory network trajectory prediction subunit predicts the operator's reference trajectory for the next 3 seconds based on historical human-machine operation temporal data, and determines the predicted reference trajectory to compensate for underwater acoustic communication delays by issuing instructions in advance.

[0075] In summary, this application achieves high-precision fusion of heterogeneous sensor data and accurate identification of operator intent by combining median filtering and wavelet decomposition with time-varying Kalman-particle filtering, and performs intent reasoning and trajectory prediction based on dynamic Bayesian networks and long short-term memory networks. It also compensates for underwater acoustic communication delay.

[0076] Please see Figure 5 , Figure 5 This is the present invention. Figure 2 A schematic diagram of an embodiment of S204. (See attached diagram.) Figure 5 As shown, the method in this embodiment may include the following steps S501-S502: S501 acquires the underwater interaction force collected by the end effector force sensor of the robotic arm.

[0077] In this embodiment, the end effector force sensor refers to a six-axis force sensor installed at the end of the robotic arm for collecting interaction forces. Underwater interaction forces refer to the forces / torques generated when the end effector of the robotic arm comes into contact with the underwater environment or the object being worked on.

[0078] In one feasible implementation, a six-axis force sensor at the end of the robotic arm collects the underwater interaction force when the robotic arm comes into contact with the underwater environment in real time, and obtains the underwater interaction force data.

[0079] The S502 transmits underwater interactive forces back to the remote control equipment via a two-way long-delay communication link to provide force feedback prompts to the operator.

[0080] In this embodiment, the bidirectional long-latency communication link refers to an uplink and downlink data transmission channel built on a deep-sea dedicated underwater acoustic modem. Force feedback cues refer to a method of transmitting underwater interaction forces back to the host computer and applying tactile force feedback to the operator via a 6-axis force feedback manipulator.

[0081] In one feasible implementation, the underwater interaction force collected by the end force sensor is transmitted back to the remote control device based on the bidirectional long-delay communication link of the underwater acoustic communication transmission unit. The human-machine interaction display subunit receives the underwater interaction force and outputs tactile force feedback to the operator, forming a force feedback prompt, so that the operator can truly perceive the underwater contact state.

[0082] In summary, this application uses an end-effector force sensor to collect underwater interactive forces and transmits them back via a two-way long-delay communication link to provide force feedback prompts. This enables the operator to have a real tactile perception of the underwater end-effector interactive forces, enhancing the sense of presence and operational safety of human-machine collaborative remote operation.

[0083] In one feasible implementation, the following is also specifically performed: The system acquires transmission data from a two-way long-delay communication link, including the predicted reference trajectory and the joint states and end effector interaction forces of the robotic arm.

[0084] In this embodiment, joint state refers to the angle and angular velocity state of each joint of the robotic arm. End-effector interaction force refers to the force / torque of the robotic arm end-effector in contact with the environment.

[0085] Based on the wave variable energy shaping model, passive constraint processing is performed on the transmitted data to obtain wave variable signals that satisfy the energy conservation constraint. The wave variable signals are used to suppress system oscillations caused by communication delay.

[0086] In this embodiment, the wave variable energy shaping model refers to a communication model constructed based on wave variable theory that ensures passivity through link energy conservation constraints. Passivity constraints are constraints that ensure the input energy of the communication link is not greater than the output energy, used to maintain system stability. Wave variable signals refer to transmission signals that satisfy energy conservation constraints after wave variable transformation processing.

[0087] The wave variable signal is transmitted bidirectionally through a bidirectional long-delay communication link.

[0088] In one feasible implementation, the underwater acoustic communication transmission unit acquires transmission data from a bidirectional long-delay communication link. This transmission data includes downlink predicted reference trajectories, human-machine control weight commands, and uplink joint states and end-effector interaction forces of the robotic arm. Based on a wave variable energy shaping model, the transmission data undergoes passive constraint processing, converting traditional power variables into wave variables to obtain wave variable signals that satisfy energy conservation constraints. This wave variable signal is then transmitted bidirectionally through the bidirectional long-delay communication link to ensure that the teleoperation system remains oscillating and stable under communication delays of 3 seconds or more.

[0089] In summary, this application uses a wave variable energy shaping model to perform passive constraint processing on transmitted data and transmit wave variable signals, ensuring energy conservation and passivity in long-delay underwater acoustic communication links, effectively suppressing system oscillations caused by communication delays, and improving the stability of the teleoperation system.

[0090] Due to the insufficient generalization ability of underwater operation skills, traditional teaching trajectories cannot adapt to varying ocean currents. Therefore, it is necessary to establish a skill learning framework enhanced with fluid disturbances for iterative transfer optimization of virtual and real-world collaborative skills. Based on this, please refer to... Figure 6 , Figure 6 This is the present invention. Figure 2 A schematic diagram of an embodiment of S204. (See attached diagram.) Figure 6 As shown, the method in this application embodiment may include the following steps S601-S604: S601, acquire samples of ocean current disturbances at multiple velocities.

[0091] In this embodiment of the application, the multi-velocity ocean current disturbance sample refers to a set of flow field disturbance data containing different flow velocities and vortex states generated by computational fluid dynamics simulation.

[0092] In one feasible implementation, the computational fluid dynamics fluid disturbance simulation module of the virtual-real simulation skills training platform simulates complex flow fields with different flow velocities from 0.3 to 2 sections and multiple vortices, generating massive amounts of disturbance data as samples of ocean current disturbances with multiple flow velocities.

[0093] S602, the data generated by underwater operations, the predicted reference trajectory and the multi-velocity ocean current disturbance samples are superimposed and amplified to obtain the amplified training samples.

[0094] In this embodiment, the overlay augmentation process refers to the process of overlaying real operational data with simulated flow field perturbation data to increase sample diversity. Augmented training samples refer to a training dataset containing rich flow field variation characteristics after perturbation overlay.

[0095] In one feasible implementation, the data generated by underwater operations, the predicted reference trajectory, and the multi-velocity ocean current disturbance samples generated by the computational fluid dynamics module are superimposed and expanded. The massive disturbance data is superimposed on the basic teaching trajectory to expand the training sample set and obtain the expanded training samples.

[0096] S603 updates and corrects the dynamic motion primitive model based on expanded training samples to generate skill trajectories.

[0097] In this embodiment, the dynamic motion primitive model refers to an improved dynamic motion primitive dynamic equation model with embedded fluid dynamics compensation terms. The skill trajectory refers to a stable operational trajectory generated based on the dynamic motion primitive model that is resistant to ocean current disturbances.

[0098] In one feasible implementation, the improved dynamic motion primitive skill learning module embeds fluid dynamics compensation terms into the traditional dynamic motion primitive dynamic equations. Based on expanded training samples, the dynamic motion primitive model is updated and corrected to generate a stable operational trajectory resistant to ocean current disturbances as the skill trajectory. Simultaneously, the virtual-real mapping correction module employs an inverse reinforcement learning algorithm to reduce the discrepancy between virtual simulation and real deep-sea working conditions, achieving adaptive skill transfer across scenarios.

[0099] S604 performs layered storage processing of skill trajectories to obtain a trajectory skill library.

[0100] In this embodiment, hierarchical storage processing refers to a management process that divides skill trajectories into different levels according to task complexity for storage. The trajectory skill library refers to a database containing skill trajectories at multiple levels.

[0101] In one feasible implementation, the hierarchical skill library stores skill trajectories in a hierarchical manner, specifically divided into three layers: basic capture, advanced detection, and expert equipment maintenance, to obtain a trajectory skill library. Existing operational skills can be quickly reused based on trajectory similarity retrieval.

[0102] Please refer to the following: Figure 7 , Figure 7 This is a schematic diagram illustrating the construction process of the trajectory skill library provided by this invention. For example... Figure 7 As shown, the process first enters the data acquisition stage, acquiring real operational data such as the joint trajectory, flow velocity, and fluid three-component torque of the robotic arm and uploading it to the simulation platform. Next, Computational Fluid Dynamics (CFD) perturbation data augmentation is performed, simulating multi-velocity and multi-vortex perturbations superimposed on the basic teaching trajectory to expand the training sample set. Then, an improved Dynamic Movement Primitive (DMP) model is trained, embedding fluid dynamics compensation terms into the traditional DMP dynamic equations to train stable trajectories resistant to flow field perturbations. It is then determined whether the task is entirely new; if not, similar skills are retrieved from the skill library based on trajectory similarity matching, and the original DMP parameters are fine-tuned. If so, a new skill category is created, and the skill trajectories are finally stored in a hierarchical skill library containing basic, advanced, and expert layers. Afterward, inverse reinforcement learning is performed to correct the virtual-real bias, adjusting the dynamic parameters of the DMP by comparing the simulation and real trajectory errors to eliminate the virtual-real bias. Finally, the skill parameters are sent to the upper-level Long Short-Term Memory (LSTM) network. The system uses Memory (LSTM) to update the trajectory generation logic of the prediction model; finally, it determines whether the next iteration is needed. If so, it returns to the data acquisition stage; if not, it evaluates whether the job quality meets the standards. If it does, the skill training ends.

[0103] Please refer to the following: Figure 8 , Figure 8 This is a schematic diagram of the architecture of the Long Short-Term Memory (LSTM) network provided by this invention. Figure 8As shown, this Long Short-Term Memory (LSTM) network uses historical trajectory sequences of the past T seconds extracted by a sliding window (step size T), 6-DOF 1kHz joint angle sequences, N-channel × 3-dimensional surface electromyography (SEM) temporal features, and end force / torque as multi-source inputs. These are mapped to d-dimensional feature vectors by the input layer and then concatenated. The concatenated d-dimensional feature vectors are then sequentially input into the n-layer hidden layers of the LSM network (including LSM network hidden layers 1 to n) for temporal feature extraction and state update. Subsequently, the network outputs two branches: one branch represents the probability of K-class distributed motion intentions through the fully connected layer, and the other branch generates a reference trajectory sequence for the next 3 seconds through the trajectory output layer, thereby simultaneously realizing the temporal reasoning of the operator's motion intentions and the generation of the predicted reference trajectory.

[0104] In summary, this application solves the problem of virtual-real discrepancy between virtual simulation and real deep-sea working conditions by superimposing and expanding perturbation samples generated through computational fluid dynamics simulation, and then storing the generated skill trajectories in a layered manner based on an improved dynamic motion primitive model and inverse reinforcement learning. This enhances the generalization ability of operational skills to varying ocean currents and enables skill reuse across current velocities and tasks.

[0105] In one feasible implementation, after the execution step of performing hierarchical storage processing of skill trajectories to obtain a trajectory skill library, the following is further specifically executed: Obtain the mission trajectory of the underwater mission to be processed.

[0106] In this embodiment of the application, the task trajectory refers to the expected motion path of the end effector of the robotic arm for the underwater task to be processed.

[0107] The task trajectory is searched and matched with the trajectory skill library to obtain the matching skill trajectory.

[0108] In this embodiment of the application, the retrieval and matching process refers to the process of searching for historical skill trajectories in the trajectory skill database that are most similar to the task trajectory based on trajectory similarity. The matched skill trajectory refers to the retrieved historical skill trajectory with the highest similarity to the task to be processed.

[0109] For example, the dynamic time-normalized distance between the task trajectory and each skill trajectory in the trajectory skill library is calculated, and the skill trajectory with the smallest distance, i.e. the highest similarity, is selected as the matching skill trajectory.

[0110] The matching skill trajectory is adaptively corrected based on a real-time fluid dynamics model to generate a target execution trajectory, which is then used to control the robotic arm to perform the underwater task to be processed.

[0111] In this embodiment, the real-time fluid dynamics model refers to a dynamic model capable of calculating the fluid feedforward compensation torque under the current flow field in real time. Adaptive correction refers to the process of torque compensation and trajectory adjustment for the matched skill trajectory based on the real-time fluid dynamics model. The target execution trajectory refers to the final execution trajectory suitable for the current real flow field environment after adaptive correction.

[0112] In one feasible implementation, the matching skill trajectory is adaptively corrected based on a real-time fluid dynamics model. The fluid feedforward compensation torque is calculated in real time in conjunction with the current flow field to correct and compensate the joint angle of the matching skill trajectory, thereby generating a target execution trajectory. This target execution trajectory is then used to control the robotic arm to perform the underwater task to be processed.

[0113] For example, when performing a new underwater mission, after retrieving a matching skill trajectory from the trajectory skill library, the matching skill trajectory is superimposed and corrected based on the fluid feedforward compensation torque solved in real time at the current moment to generate the target execution trajectory and complete the non-repetitive teaching operation.

[0114] In summary, this application achieves rapid matching and adaptive correction of existing skills in a new flow field environment by using trajectory similarity retrieval to match historical skills and combining real-time fluid dynamics models to adaptively correct the generated target execution trajectory. This avoids repeated teaching and improves the flexibility and execution efficiency of underwater operations.

[0115] The following will combine Figure 9 This application provides a detailed description of the underwater robot control device provided in its embodiments. It should be noted that... Figure 9 The underwater robot control device in the present application is used to perform the functions described herein. Figures 1-6 The methods shown in the embodiments are illustrated for ease of explanation, showing only the parts relevant to the embodiments of this application. For specific technical details not disclosed, please refer to this application. Figures 1-6 In the embodiment shown, the underwater robot control device 900 may include a data prediction unit 901, a compensation torque acquisition unit 902, a total driving torque acquisition unit 903, and a work execution unit 904, as detailed below: The data prediction unit 901 is used to acquire multimodal perception data generated by the operator when operating the remote control device, perform data prediction on the multimodal perception data, and determine the operator's motion intention and predict the reference trajectory. The compensation torque acquisition unit 902 is used to acquire the link geometry parameters of the robotic arm, perform fluid dynamics calculation on the link geometry parameters, and obtain the fluid feedforward compensation torque. The fluid feedforward compensation torque is the total fluid dynamic load including viscous drag torque, additional mass torque and separately solved buoyancy torque. The total driving torque acquisition unit 903 is used to acquire the rigid body foundation control torque of the robotic arm, and to superimpose the rigid body foundation control torque with the fluid feedforward compensation torque to obtain the total driving torque. The work execution unit 904 is used to drive the robotic arm to perform trajectory tracking based on the total driving torque in order to perform underwater operations.

[0116] The underwater robot control device 900 provided in the above embodiments can realize the technical solutions described in the above underwater robot control method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above underwater robot control method embodiments, and will not be repeated here.

[0117] like Figure 10 As shown, the present invention also provides an underwater robot control system 10. The underwater robot control system 10 includes a processor 11, a memory 12, a display 13, a communication module 14, a sensing component 15, and a robotic arm drive component 16. Figure 10 Only some components of the underwater robot control system 10 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0118] In some embodiments, processor 11 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 12 or process data, such as the underwater robot control method of the present invention.

[0119] In some embodiments, the memory 12 may be an internal storage unit of the underwater robot control system 10, such as a hard disk or memory of the underwater robot control system 10. In other embodiments, the memory 12 may also be an external storage device of the underwater robot control system 10, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the underwater robot control system 10.

[0120] Furthermore, the memory 12 may include both internal storage units of the underwater robot control system 10 and external storage devices. The memory 12 is used to store application software and various types of data installed on the underwater robot control system 10.

[0121] Display 13, in some embodiments, can be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. Display 13 is used to display information from the underwater robot control system 10 and to display a visual user interface. Communication module 14 is used to receive multimodal perception data from the operator via a bidirectional long-latency communication link and to transmit the joint states and end-effector interaction forces of the robotic arm. Sensing component 15 is used to acquire the link geometry parameters and end-effector interaction forces of the robotic arm. Robotic arm drive component 16 is used to drive the robotic arm to perform trajectory tracking for underwater operations under the control of the processor. Components 11-16 of the underwater robot control system 10 communicate with each other via a system bus.

[0122] In one embodiment, when the processor 11 executes the underwater robot control program in the memory 12, the following steps can be performed: Acquire multimodal sensing data generated by the operator when operating the remote control device, perform data prediction on the multimodal sensing data, determine the operator's motion intention and predict reference trajectory; Obtain the link geometry parameters of the robotic arm, perform fluid dynamics calculations on the link geometry parameters, and obtain the fluid feedforward compensation torque. The fluid feedforward compensation torque is the total fluid dynamic load including viscous drag torque, additional mass torque, and separately solved buoyancy torque. Obtain the rigid body foundation control torque of the robotic arm, and superimpose the rigid body foundation control torque with the fluid feedforward compensation torque to obtain the total driving torque; The total driving torque drives the robotic arm to perform trajectory tracking in order to carry out underwater operations.

[0123] It should be understood that when the processor 11 executes the underwater robot control program in the memory 12, in addition to the functions mentioned above, it can also perform other functions, as can be found in the description of the corresponding method embodiments above.

[0124] Accordingly, this application also provides a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, they can implement the steps or functions of the underwater robot control methods provided in the above-described method embodiments.

[0125] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0126] The underwater robot control method, device, and control system provided by this invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A control method for an underwater robot, characterized in that, This technology is applied to underwater robots, which are equipped with robotic arms for performing underwater operations. The underwater robot is connected to a remotely operated device via a bidirectional long-latency communication link. The remotely operated device is used by an operator to manually control the robotic arm, including: The system acquires multimodal sensing data generated by the operator when operating the remote control device, performs data prediction on the multimodal sensing data, and determines the operator's motion intention and predicted reference trajectory. Obtain the link geometry parameters of the robotic arm, perform fluid dynamics calculation on the link geometry parameters to obtain the fluid feedforward compensation torque, which is the total fluid dynamic load including viscous drag torque, additional mass torque and separately solved buoyancy torque; The rigid body base control torque of the robotic arm is obtained, and the rigid body base control torque is superimposed with the fluid feedforward compensation torque to obtain the total driving torque. The robotic arm is driven by the total driving torque to perform trajectory tracking in order to carry out underwater operations.

2. The method according to claim 1, characterized in that, The step of performing hydrodynamic calculations on the connecting rod's geometric parameters to obtain the fluid feedforward compensation torque includes: The homogeneous transformation matrix of the connecting rod's geometric parameters is solved to obtain the Jacobian matrix of the connecting rod's centroid. Based on the Morrison equation, the Jacobian matrix of the center of mass of the connecting rod is integrated to obtain the viscous drag torque and the additional mass torque. The buoyancy of a single link is mapped based on the Jacobian matrix of the link's center of mass to obtain the separately solved buoyancy torque; The fluid feedforward compensation torque is obtained by superimposing the viscous drag torque, the additional mass torque, and the separately solved buoyancy torque.

3. The method according to claim 1, characterized in that, The step of performing data prediction on the multimodal sensing data to determine the operator's motion intention and predicted reference trajectory includes: The multimodal sensing data is subjected to composite filtering to obtain noise-reduced sensing data; Based on the noise reduction sensing data, spatiotemporal registration and fusion processing is performed to obtain unified spatiotemporal aligned fusion features; Intent-time reasoning and trajectory prediction are performed on the unified spatiotemporal alignment fusion features to determine the operator's motion intent and predicted reference trajectory. The predicted reference trajectory is a future trajectory used to compensate for the latency of the bidirectional long-delay communication link.

4. The method according to claim 3, characterized in that, Before performing composite filtering on the multimodal sensing data to obtain noise-reduced sensing data, the method further includes: Acquire the amplitude of the operator's electromyographic signal; The fatigue state value is obtained by performing fatigue measurement based on the amplitude of the electromyographic signal. When the fatigue state value exceeds a preset threshold, a smooth switching process for control weights is performed to obtain a dynamic human-machine hybrid control weight instruction. The dynamic human-machine hybrid control weight instruction is an instruction used to adjust the allocation ratio between the operator's manual control and the underwater robot's autonomous control.

5. The method according to claim 1, characterized in that, After driving the robotic arm to perform trajectory tracking based on the total driving torque to execute a preset underwater operation, the method further includes: The underwater interaction force collected by the end effector force sensor of the robotic arm is acquired; The underwater interactive force is transmitted back to the remote control device via the bidirectional long-delay communication link to provide force feedback prompts to the operator.

6. The method according to claim 1, characterized in that, The method further includes: The transmission data of the bidirectional long-delay communication link is acquired, and the transmission data includes the predicted reference trajectory and the joint state and end effector interaction force of the robotic arm; The transmitted data is subjected to passive constraint processing based on the wave variable energy shaping model to obtain a wave variable signal that satisfies the energy conservation constraint. The wave variable signal is used to suppress system oscillations caused by communication delay. The wave variable signal is transmitted bidirectionally through the bidirectional long-delay communication link.

7. The method according to claim 1, characterized in that, The method further includes: Acquire samples of ocean current disturbances at multiple speeds; The data generated from underwater operations, the predicted reference trajectory, and the multi-velocity ocean current disturbance samples are superimposed and amplified to obtain expanded training samples. The dynamic motion primitive model is updated and corrected based on the expanded training samples to generate a skill trajectory. The dynamic motion primitive model is a model used to generate a stable operation trajectory that is resistant to ocean current disturbances. The skill trajectories are stored in layers to obtain a trajectory skill library.

8. The method according to claim 7, characterized in that, After performing hierarchical storage processing on the skill trajectories to obtain a trajectory skill library, the method further includes: Obtain the mission trajectory of the underwater mission to be processed; The task trajectory is searched and matched with the trajectory skill library to obtain the matched skill trajectory; The matching skill trajectory is adaptively corrected based on a real-time fluid dynamics model to generate a target execution trajectory, which is then used to control the robotic arm to perform the underwater task to be processed.

9. An underwater robot control device, characterized in that, This technology is applied to underwater robots, which are equipped with robotic arms for performing underwater operations. The underwater robot is connected to a remotely operated device via a bidirectional long-latency communication link. The remotely operated device is used by an operator to manually control the robotic arm, including: The data prediction unit is used to acquire multimodal perception data generated by the operator when operating the remote control device, perform data prediction on the multimodal perception data, and determine the operator's motion intention and prediction reference trajectory. The compensation torque acquisition unit is used to acquire the link geometry parameters of the robotic arm, perform fluid dynamics calculation on the link geometry parameters, and obtain the fluid feedforward compensation torque. The fluid feedforward compensation torque is the total fluid dynamic load including viscous drag torque, additional mass torque and separately solved buoyancy torque. The total driving torque acquisition unit is used to acquire the rigid body foundation control torque of the robotic arm, and to superimpose the rigid body foundation control torque with the fluid feedforward compensation torque to obtain the total driving torque. The operation execution unit is used to drive the robotic arm to perform trajectory tracking in order to perform underwater operations based on the total driving torque.

10. An underwater robot control system, characterized in that, It includes a communication module, sensing components, robotic arm drive components, memory, and a processor, among which, The communication module is used to receive multimodal perception data from the operator through a two-way long-latency communication link, and to send the joint state and end effector interaction force of the robotic arm. The sensing components are used to acquire the link geometry parameters and end effector force of the robotic arm. The robotic arm drive assembly is used to drive the robotic arm to perform trajectory tracking under the control of the processor to carry out underwater operations. The memory is used to store executable program code; The processor is configured to perform data prediction on the multimodal sensing data, determine the operator's motion intention and predict a reference trajectory, wherein the predicted reference trajectory is a future trajectory used to compensate for the latency of the bidirectional long-delay communication link; perform hydrodynamic calculations on the linkage geometric parameters to obtain a fluid feedforward compensation torque; acquire the rigid body base control torque of the robotic arm, and superimpose the rigid body base control torque with the fluid feedforward compensation torque to obtain a total driving torque; and control the robotic arm drive assembly to drive the robotic arm to perform trajectory tracking to execute a preset underwater operation based on the total driving torque.

Images