River channel floating object grabbing track control method oriented to three-axis freedom degree grabbing

Abstract

The invention relates to the technical field of track motion control, and discloses a three-axis degree-of-freedom grabbing-oriented channel floating object grabbing track control method, which comprises the following steps of: firstly, fusing sensor data to obtain a target and a mechanism motion state; calculating a critical response period reflecting the response hysteresis degree according to the transmission singularity index of the mechanism and the tail end equivalent inertia; then, dynamically adjusting a sampling interval of a rolling programming controller by utilizing the generated prediction step length adjustment coefficient, and constructing a dynamic model containing a structure flexible vibration state in the controller to prevent time domain folding of a prediction window; and finally, establishing a multi-target cost function containing a time domain re-calibration weight, generating a driving instruction through continuous numerical iteration, and adaptively correcting model parameters based on the monitored structural vibration energy. The dynamic degradation problem of a parallel mechanism in a near singular area can be effectively solved, and vibration suppression and energy consumption optimization in the high-speed dynamic grabbing process are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of trajectory motion control, more particularly, it relates to a river floating object grabbing trajectory control method for three-axis freedom grabbing. BACKGROUND

[0002] With the increasing demand for water environmental protection, river floating object cleaning gradually shifts from manual salvage to automated operation. In this process, three-degree-of-freedom parallel mechanisms (such as Delta mechanism) are often used for high-speed dynamic pick-and-place tasks due to their lightweight structure, fast motion, and ability to achieve pure translation at the end. In order to achieve accurate grabbing of floating objects on flowing water surface, existing control schemes usually use advanced algorithms such as model predictive control (MPC) to handle trajectory tracking problems under multiple constraints through online rolling optimization. Such control schemes can maintain good tracking effect in the center area of the mechanism workspace under normal working conditions, with fixed prediction step and control period.

[0003] However, in complex scenarios such as river cleaning, the target object often drifts with the water flow to the edge area of the robot workspace. In these areas, three-degree-of-freedom parallel mechanisms face severe structural dynamics challenges, and existing technologies have the following significant deficiencies: When the parallel mechanism approaches the boundary of the workspace (near-singularity area), the kinematic transmission relationship will change dramatically. At this time, the equivalent stiffness of the mechanism along a certain direction will significantly decrease, while the equivalent inertia reflected to the end will sharply increase. Existing control algorithms usually consider the system as a rigid body or a parameter slowly varying system, ignoring the dramatic change in dynamics caused by changes in position and attitude, resulting in that the instructions issued by the controller cannot be effectively responded by the mechanism, causing serious execution lag.

[0004] The mainstream MPC algorithm usually adopts a fixed prediction time window. When the mechanism enters the near-singularity area, the natural vibration frequency of the system (especially the dominant low-frequency mode) will significantly decrease, resulting in a longer actual response period of the system. If the length of the prediction window is fixed and shorter than the response period of the system, the controller will not be able to observe the complete state evolution process (i.e., time domain folding phenomenon occurs), thus calculating the wrong control amount based on incomplete information, causing system overshoot or even oscillation.

[0005] In high-speed reciprocating motion, the lightweight rods of parallel mechanism are easy to excite multi-modal residual vibration. The existing technology often filters these vibrations as external disturbances or noise, rather than incorporating them as inherent states of the system into the planning model. In addition, existing solutions often only consider speed factors when optimizing energy consumption, without considering the coupling effect of pose changes on joint torque. This makes it difficult to balance between suppressing vibration and reducing energy consumption, especially in the limit position where the mechanism output is difficult and the stiffness is soft. Forced high-gain control not only fails to eliminate errors, but also leads to motor overheating and structural resonance. SUMMARY

[0006] The present application provides a river floating object grabbing trajectory control method for three-axis freedom grabbing, which solves the technical problems proposed in the background art.

[0007] The present application provides a river floating object grabbing trajectory control method for three-axis freedom grabbing, which includes: Collect and fuse the position and velocity data of the floating object target and the current motion state data of the three-degree-of-freedom parallel mechanism end through sensors; According to the geometric relationship of the rods of the three-degree-of-freedom parallel mechanism, the transmission singularity index representing the transmission capacity is calculated, and the equivalent inertia of the end along the minimum constraint direction is obtained by combining the rotor inertia of the drive motor; According to the transmission singularity index and the equivalent inertia of the end, a critical response period reflecting the degree of dynamic response lag of the mechanism is established, and a prediction step adjustment coefficient is generated based on the critical response period; The prediction step adjustment coefficient is used to dynamically adjust the prediction sampling interval of the rolling planning controller, and a dynamics model containing the structural flexible vibration state is constructed in the rolling planning controller to cover the flexible vibration period under a fixed prediction step number; A multi-objective cost function containing a weight based on the prediction step adjustment coefficient is established, and the drive acceleration command is generated by continuous numerical iteration and driven to execute; The structural vibration energy of the three-degree-of-freedom parallel mechanism end in the minimum constraint direction is monitored, and the model correction parameters used to calculate the critical response period are adaptively corrected based on the structural vibration energy.

[0008] The beneficial effects of the present application are: the present application solves the control failure problem of three-degree-of-freedom parallel mechanism in the near-singular region due to slow dynamics response by constructing a critical response period closely related to the mechanism posture. The present application generates a prediction step adjustment coefficient using the transmission singularity index and the equivalent inertia at the end, dynamically stretches the time scale of the rolling planning, ensures that the prediction window can cover the flexible vibration period of the system under any working condition, and effectively avoids the control divergence caused by time domain folding; at the same time, by injecting the structure flexibility state into the dynamics model and driving the adaptive update with the structure vibration energy, the active suppression of residual vibration and energy consumption collaborative optimization are realized, which significantly improves the dynamic tracking accuracy and operation stability of the robot in the extreme edge working conditions such as river cleaning. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 is a flow chart of the river floating object grabbing trajectory control method for three-axis degree-of-freedom grabbing of the present application. DETAILED DESCRIPTION

[0010] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is merely meant to provide a better understanding of the subject matter described herein and can include changes, modifications, additions or omissions of the functions and arrangements of the elements discussed without departing from the scope of the present description. Various examples can omit, substitute or add various procedures or components as appropriate, and the specific details of various implementations can be adapted to the context in which they are used. Also, features described with respect to some examples can be combined in other examples.

[0011] It should be noted that the technical terms or scientific terms used by one or more embodiments of the present application should be understood as the usual meaning understood by those skilled in the art in the field to which the present application belongs, unless otherwise defined. The terms "first", "second" and the like used in one or more embodiments of the present application do not represent any order, number or importance, but are only used to distinguish different components. The terms "include" or "contain" and the like mean that the elements or objects appearing before the terms encompass the elements or objects listed after the terms and their equivalents, and do not exclude other elements or objects. The terms "connected" or "connected" and the like are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms "up", "down", "left", "right" and the like are only used to represent relative positional relationships, and when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0012] As shown in Figure 1 The river floating object grabbing trajectory control method for three-axis degree-of-freedom grabbing includes: The position and velocity data of the floating target are collected by sensors and fused with the current motion state data of the end effector of the three-degree-of-freedom parallel mechanism; The transmission singularity index characterizing the transmission capacity is calculated based on the geometric relationship of the links of the three-degree-of-freedom parallel mechanism, and the equivalent inertia at the end along the minimum constraint direction is obtained by combining the rotor inertia mapping of the drive motor. A critical response period reflecting the degree of hysteresis in the dynamic response of the mechanism is established based on the transmission singularity index and the end-effector equivalent inertia, and a predictive step size adjustment coefficient is generated based on the critical response period. The prediction sampling interval of the rolling planning controller is dynamically adjusted using the prediction step size adjustment coefficient, and a dynamic model containing the flexible vibration state of the structure is constructed within the rolling planning controller so that the prediction window covers the flexible vibration period under a fixed number of prediction steps. A multi-objective cost function is established, which includes weights based on the predicted step size adjustment coefficient. The driving acceleration command is generated and executed by continuous numerical iteration. The structural vibration energy at the end of the three-degree-of-freedom parallel mechanism in the direction of minimum constraint is monitored, and the model correction parameters used to calculate the critical response period are adaptively corrected based on the structural vibration energy.

[0013] Preferably, the position and velocity data of the floating target are collected and fused with the current motion state data of the end effector of the three-degree-of-freedom parallel mechanism by sensors, including: Step A, Time Alignment: Obtain measurement data at the target measurement time, and align the measurement data to the current control time based on the time difference between the current control time and the target measurement time using a first-order extrapolation formula; Step B, coordinate transformation: using the calibrated rotation matrix and translation vector, the aligned measurement data is transformed from the sensor coordinate system to the target base coordinate system to obtain the current measurement value; Step C, Exponential Fusion: The position and velocity data of the floating target and the current motion state data of the end effector of the three-degree-of-freedom parallel mechanism are calculated using the following first-order exponential fusion formula: in, The formula for calculating the fixed fusion coefficient is as follows: In the formula, , These are the position fusion value and velocity fusion value for the current control cycle, respectively. , These are the position fusion value and velocity fusion value from the previous control cycle, respectively. , These are the current position measurement value and the current velocity measurement value after transformation in step B, respectively; The sampling period; This is the equivalent filtering time constant; It is an exponential function with the natural constant as its base.

[0014] The measured position of a floating object is a physical quantity that reflects the spatial position of the floating object in the sensor coordinate system. It can be collected by a visual sensor or a lidar.

[0015] The velocity measurement of a floating target is a physical quantity that reflects the speed and direction of the floating object's movement. It can be collected through inter-frame difference analysis of visual sensors or velocity calculation by lidar.

[0016] The end position measurement of a three-degree-of-freedom parallel mechanism is a physical quantity that reflects the spatial position of the actuator end in the sensor coordinate system. It can be acquired by an encoder combined with forward kinematics or a vision sensor.

[0017] The end-effector velocity measurement value of a three-degree-of-freedom parallel mechanism is a physical quantity that reflects the speed and direction of the actuator's end motion. It can be acquired through encoder differential or visual sensor inter-frame analysis.

[0018] The timestamp of sensor measurement data records the time information at the moment the sensor collects data, and can be collected through the sensor's built-in clock module.

[0019] The current control moment is the specific time point at which the control system executes this data processing and instruction generation. It is preferably a millisecond-level time for unified system timing to ensure the consistency of the control cycle.

[0020] Rigid body transformation parameters are rotation matrices and translation vectors that realize the transformation between the sensor coordinate system and the base coordinate system. They can be acquired through calibration experiments using a calibration plate.

[0021] The system sampling period is the time interval between the control system repeatedly executing data acquisition and command output, preferably 1 millisecond or 2 milliseconds, to meet the balance requirements of real-time performance and computational load in high-speed scene capture.

[0022] The equivalent filtering time constant is a parameter that adjusts the smoothness of exponential fusion, preferably between 5 and 20 milliseconds, depending on the sensor noise characteristics; the higher the noise, the larger the value.

[0023] The minimum value for preventing division by zero is a very small value that avoids division by zero errors in calculations. It is preferably 1e-6 so as not to affect the normal calculation results and to effectively avoid numerical anomalies.

[0024] The combined strategy of recent sample-and-hold and first-order linear extrapolation includes: firstly, setting a time difference threshold of 0.5 milliseconds; when the time difference between the current control time and the measurement time is less than this threshold, first-order linear extrapolation is used, that is, the current time data is extrapolated based on the measurement data and the time difference; when the time difference is greater than or equal to this threshold, recent sample-and-hold is used, that is, the most recent measurement data is directly used. For example, if the measurement time data is position 10 cm, velocity 5 cm / s, and time difference 0.3 milliseconds, the extrapolated current position is 10 + 5 × 0.3 = 11.5 cm.

[0025] First-order exponential fusion includes a fusion coefficient calculated using a formula based on the sampling period and the equivalent filtering time constant. For example, with a sampling period of 1 millisecond and an equivalent filtering time constant of 10 milliseconds, the fusion coefficient is approximately 0.095. This algorithm combines historical data and current measurement data through weighted summation. The weight of historical data is 1 minus the fusion coefficient, and the weight of current data is the fusion coefficient itself. This approach preserves the stability of historical data while rapidly responding to changes in current data.

[0026] Rigid body transformation parameter calibration includes: using a checkerboard calibration plate, fixing the calibration plate at a known position in the base coordinate system, acquiring images of the calibration plate from different angles using sensors, calculating the rotation matrix and translation vector using a camera calibration algorithm, and repeatedly acquiring 10 sets of data and taking the average value to ensure transformation accuracy.

[0027] The selection of the equivalent filtering time constant includes: 5 milliseconds when using a low-noise LiDAR; and 15 to 20 milliseconds when using a high-noise ordinary camera, balancing response speed and noise resistance.

[0028] Initialization method during startup: When the system starts up, the measurement data acquired for the first time and after coordinate transformation is directly used as the historical state data of the previous control cycle. That is, when k=0, the position and velocity of the previous cycle are equal to the position and velocity measured for the first time.

[0029] The time alignment threshold is uniformly set to 0.5 milliseconds. This value is determined based on the sampling frequency and transmission delay of common sensors and can cover the synchronization requirements in most scenarios.

[0030] Preferably, the transmission singularity index characterizing the transmission capacity is calculated based on the geometric relationship of the links in the three-degree-of-freedom parallel mechanism, and the equivalent inertia at the end along the minimum constraint direction is obtained by combining the rotor inertia mapping of the drive motor, including: Step A, Singular Value Decomposition: Based on the current drive joint position Calculation speed Jacobian matrix And perform the following singular value decomposition on it: Step B, Index Calculation: Calculate the transmission singularity index. And extract the minimum constraint direction. : Step C, Inertia Mapping: Constructing the equivalent inertia matrix on the drive side The pseudo-inverse of the velocity Jacobian matrix is ​​calculated using the damped least squares method. Then, the platform reflection inertia matrix is ​​calculated. : Step D, Inertia Projection: Project the platform reflection inertia matrix onto the minimum constraint direction to obtain the end-effector equivalent inertia. : In the formula, , , These are the maximum singular value, intermediate singular value, and minimum singular value of the velocity Jacobian matrix, respectively. To prevent the removal of minute quantities; It is a left singular vector matrix; express The third column vector; It is a diagonal matrix that includes the inertia of the motor rotor, reducer, and connecting rod. It is the identity matrix; The adaptive damping coefficient varies with the minimum singularity; Indicates matrix transpose; This represents finding the inverse of a matrix.

[0031] The position of the drive joint is a physical quantity that reflects the rotation angle of the drive joint in a three-degree-of-freedom parallel mechanism, and can be directly acquired by a joint encoder.

[0032] The geometric parameters of the links in a three-degree-of-freedom parallel mechanism are key parameters for defining the structural dimensions of the mechanism, including arm length, base radius, platform radius, etc. The preferred values ​​are determined based on the actual mechanism design to meet the working space requirements for grasping floating objects in the river.

[0033] The moment of inertia of a motor rotor is a physical quantity representing the rotational inertia of the motor rotor itself. It can be obtained by querying data tables provided by the motor manufacturer.

[0034] The moment of inertia of a speed reducer is a physical quantity representing the rotational inertia of the transmission components inside the speed reducer. It can be obtained by looking up data provided by the speed reducer manufacturer.

[0035] The equivalent inertia of a link is the rotational inertia that equates the mass of the link to the drive joint side. It can be acquired through sinusoidal velocity excitation experiments combined with least squares fitting.

[0036] The adaptive damping coefficient is an adjustment parameter used to construct the damping least squares inverse matrix. It is preferred to have an initial value of 0.002 and a dynamic adjustment range of 0.002 to 0.05, so that it increases as the minimum singular value decreases, thus avoiding numerical divergence when near singularity.

[0037] The minimum value for preventing division by zero is an extremely small value that avoids division by zero errors when calculating the singularity index of the transmission. It is preferably 1e-12 so as not to affect the normal calculation results and to effectively avoid numerical anomalies.

[0038] Continuous sign alignment processing includes: Feature direction vectors obtained from singular value decomposition may exhibit sign jumps. By introducing an alignment coefficient, the dot product of the current feature direction vector and the previous cycle's feature direction vector is calculated. The dot product is then substituted into the hyperbolic tangent function to obtain the sign alignment coefficient. This coefficient is used to correct the current feature direction vector before normalization. For example, if the previous cycle's feature direction vector is 0.8 0.5 0.3, and the current original vector is -0.78 0.52 0.31, the dot product is approximately 0.98, and the alignment coefficient is close to 1. After correction, the vector remains consistent with the previous cycle's direction, avoiding control fluctuations caused by abrupt sign changes.

[0039] The construction of the damping least squares inverse matrix and the inertia mapping include: constructing a damping least squares inverse matrix based on the transpose of the velocity Jacobian matrix and combining it with adaptive damping coefficients. This matrix retains the mapping relationship of the Jacobian matrix and improves the numerical stability in near-singular conditions through the damping term. Multiplying the diagonal inertia matrix of the drive side with this inverse matrix and its transpose achieves the mapping of inertia from the drive side to the end-effector platform side, resulting in the platform reflection inertia matrix reflecting the end-effector dynamics. This process considers the influence of the mechanism's kinematic transmission relationship on the inertia.

[0040] The platform reflection inertia matrix projection scalarization involves multiplying the platform reflection inertia matrix by the minimum constraint direction vector and its transpose to extract the equivalent inertia scalar along the minimum constraint direction. This method focuses on the weakest dynamic direction of the mechanism, simplifies the complexity of subsequent dynamic modeling, and makes the model more closely resemble the actual dynamic characteristics of the near-singular region.

[0041] The geometric parameters of the rods include: arm length, which is the straight-line distance between the hinge points at both ends of the active arm; base radius, which is the radius of the circle on the base where the hinge points are distributed; and platform radius, which is the radius of the circle on the end platform where the hinge points are distributed. All of these parameters are based on the dimensions marked on the mechanism design drawings.

[0042] The identification of the equivalent inertia of the link includes: applying a sinusoidal speed command with an amplitude of 0.1 radians per second and a frequency of 5 Hz to a single drive joint within a safe space, measuring the joint output torque, ignoring viscous friction in the low-speed region, obtaining the equivalent inertia of the drive side by least squares fitting through the ratio of torque to angular acceleration, and then subtracting the inertia of the motor rotor and the reducer to obtain the equivalent inertia of the link.

[0043] The adaptive damping coefficient adjustment includes: the adaptive damping coefficient is equal to the basic damping coefficient plus the enhanced damping coefficient multiplied by the smoothed logarithmic exponential function. The input of the smoothed logarithmic exponential function is the reference minimum singular value minus the current minimum singular value. The basic damping coefficient is 0.002, the enhanced damping coefficient is 0.05, and the smoothing parameter is 50.

[0044] The measures include preventing the division by zero by a small amount, which is much smaller than the normal range of the minimum singular value. Even if the minimum singular value is close to zero, adding this small amount will not significantly change the calculation result of the transmission singularity index, and at the same time effectively avoid division by zero errors.

[0045] The value of the continuous sign alignment coefficient includes: the smoothing parameter of the alignment coefficient is preferably 100. This value can make the alignment coefficient quickly approach 1 when the dot product is close to 1, and quickly approach -1 when the dot product is close to -1, so as to ensure the continuity of the feature direction vector sign.

[0046] Preferably, a critical response period reflecting the degree of dynamic response hysteresis of the mechanism is established based on the transmission singularity index and the equivalent inertia of the end effector, and a predictive step size adjustment coefficient is generated based on the critical response period, including: Step A, Stiffness Modeling: Based on the minimum singular value obtained during the calculation of the transmission singularity index. The surrogate stiffness is constructed using the following smooth attenuation formula. : Step B, Frequency Calculation: Based on the surrogate stiffness The terminal equivalent inertia and the model correction parameters Calculate the equivalent flexible vibration frequency : Step C, Period Conversion: Calculate the critical response period. : Step D, Coefficient Generation: Based on a preset reference time scale Calculate the prediction step size adjustment coefficient. : In the formula, This is the lower bound constant for stiffness; This is the nominal stiffness constant; The reference minimum singular value constant; Pi; the model correction parameters The initial value is 1, and it is updated in each control cycle by the adaptive correction step.

[0047] The lower bound constant of stiffness is a parameter that limits the minimum value of the surrogate stiffness. It is preferably 0.05 times the nominal stiffness constant to prevent the surrogate stiffness from approaching zero and causing numerical anomalies when it is near singular.

[0048] The nominal stiffness constant is a parameter characterizing the constraint stiffness of a mechanism under normal operating conditions. It is preferably 8000 N / m, which is the basic stiffness of the mechanism obtained through static stiffness calibration experiments.

[0049] The reference minimum singular value is a benchmark parameter used for standardized surrogate stiffness calculations, preferably 0.05, representing the typical range of minimum singular values ​​at the nominal operating point of the mechanism.

[0050] The reference time scale is a base time parameter used to calculate the prediction step size adjustment coefficient. It is preferably the prediction step number multiplied by the system sampling period and then divided by 2, so that the reference time scale is close to the slow mode half-cycle under normal operating conditions.

[0051] The model calibration parameters are adjustment parameters that correct the equivalent flexible vibration frequency. The initial value is preferably 1, so that the dynamic model is assumed to have no deviation in the initial state, and subsequent corrections are made through adaptive updates.

[0052] The minimum singular value is the smallest eigenvalue of the velocity Jacobian matrix, reflecting the weakness of the mechanism's transmission capability.

[0053] The end effector inertia is the platform's equivalent inertia along the direction of minimum constraint, reflecting the dynamic response characteristics of the mechanism's end effector.

[0054] The construction of the surrogate stiffness through continuous nonlinear decay includes: weighting the difference between the nominal stiffness and the lower bound of stiffness based on the ratio of the square of the minimum singular value to the sum of the squares of the reference minimum singular values, and then superimposing the lower bound of stiffness. For example, with a nominal stiffness of 8000 N / m, a lower bound of stiffness of 400 N / m, and a reference minimum singular value of 0.05, when the current minimum singular value is 0.025, the surrogate stiffness is 400 + (8000 - 400) × (0.025). 2 / (0.025) 2 +0.05 2 =400+7600×0.2=1920 N / m, achieving a smooth attenuation as the minimum singular value decreases, which fits the actual characteristics of stiffness degradation near the singular region.

[0055] The relationship between the critical response period and the prediction step size adjustment coefficient includes: the critical response period is determined by the ratio of pi to the equivalent flexible vibration frequency, directly reflecting the time scale of the slow mode; the prediction step size adjustment coefficient is obtained by squared the ratio of the critical response period to the reference time scale and adding 1, achieving adaptive scaling of the time scale. For example, with a critical response period of 0.1 seconds and a reference time scale of 0.01 seconds, the adjustment coefficient is 1 + (0.1 / 0.01). 2 =101, which reduces the prediction sampling interval to 1 / 101 of the original sampling period, thus adapting to slow mode characteristics.

[0056] The calibration of the nominal stiffness constant includes: using the static calibration method, fixing the end platform along the direction of minimum constraint, applying different levels of static tension using a tension gauge, recording the corresponding displacement data, calculating the static stiffness by the ratio of tension to displacement, and taking the average value of multiple measurements as the nominal stiffness constant.

[0057] The value of the lower bound constant of stiffness includes: selecting 0.05 times the nominal stiffness constant, which ensures that the surrogate stiffness in the near-singular region is not too small to cause numerical divergence, and also reflects the trend of stiffness decay, avoiding excessive deviation from the actual mechanism characteristics.

[0058] The selection of the reference minimum singular value includes: the minimum singular value of the calculated velocity Jacobian matrix at the nominal working point (non-edge, non-near-singular region) in the mechanism's workspace, and using this value as the reference minimum singular value to ensure the rationality of the standardized calculation.

[0059] The reference time scale is determined as follows: if the prediction step is 20 and the system sampling period is 1 millisecond, then the reference time scale is 20 × 1 millisecond / 2 = 10 milliseconds, so that the reference time scale is close to the half-cycle of the slow mode under normal operating conditions, ensuring that the scaling range of the adjustment coefficient is reasonable.

[0060] The initial values ​​of the model calibration parameters are set as follows: In the initial state, the deviation between the theoretical value of the dynamic model and the actual mechanism characteristics is unknown. The initial value is set to 1, which means that the default model has no deviation. Subsequently, the model is dynamically adjusted according to the vibration energy through adaptive updates to ensure the stability of the initial control.

[0061] Preferably, the prediction sampling interval of the rolling planning controller is dynamically adjusted using the prediction step size adjustment coefficient, and a dynamic model including the flexible vibration state of the structure is constructed within the rolling planning controller to ensure that the prediction window covers the flexible vibration period under a fixed number of prediction steps, including: Step A, Step Size Recalibration: Calculate the predicted sampling interval : Step B, parameter freezing: in the rolling planning controller... Within the prediction window of the next control cycle, the prediction step is set. The frequency and direction parameters remain unchanged: Step C, Flexible State Recursion: Construct a discrete recursive model of the flexible vibration state of the structure based on the semi-implicit Euler method: Step D, Rigid Body State Recursion: Constructing the Rigid Body Position With rigid body velocity Discrete recursive model: In the formula, The system sampling period; The prediction step size adjustment coefficient; The equivalent flexible vibration frequency; The direction of the minimum constraint; , Prediction steps Modal displacements and modal velocities; The structural damping ratio constant; The modal controlled gain constant; For prediction steps The drive acceleration command; This represents the transpose of a vector.

[0062] The structural damping ratio constant is a parameter describing the vibration attenuation characteristics of a structure. It is preferably between 0.05 and 0.15, which is the typical damping range of a lightly damped parallel mechanism, balancing vibration suppression effect and response speed.

[0063] The modal controlled gain constant is a coefficient that converts the projection of the driving acceleration into modal excitation. It is preferably 1.0, based on the dynamic matching relationship under the initial nominal state, and can be subsequently identified and corrected through experiments.

[0064] The prediction step count is the number of prediction steps optimized by the rolling planning controller in one operation, preferably 20, to balance the engineering experience of control real-time performance with prediction window coverage.

[0065] The system sampling period is the time interval at which the control system repeatedly performs data processing.

[0066] The prediction step size adjustment factor is a scaling factor that dynamically adjusts the prediction sampling interval.

[0067] The equivalent flexible vibration frequency is a frequency parameter that reflects the slow-mode vibration characteristics of a structure.

[0068] The direction of minimum constraint is the spatial direction in which the transmission capability of the mechanism is weakest.

[0069] The prediction sampling interval is the time interval between each prediction step in the rolling planning.

[0070] Modal displacement is the displacement state of a structure's flexible vibration in the direction of minimum constraint.

[0071] Modal velocity is the velocity state of a structure's flexible vibration in the direction of minimum constraint.

[0072] The drive acceleration command is the end acceleration control quantity output by the controller.

[0073] Rigid body position refers to the rigid motion position state of the end effector.

[0074] Rigid body velocity is the rigid motion velocity state of the end effector.

[0075] The predicted state variable is a comprehensive predicted state that integrates rigid body motion and flexible vibration.

[0076] Prediction window parameter freezing includes fixing the equivalent flexible vibration frequency and minimum constraint direction to the calculated values ​​of the current cycle within the prediction window of the current control cycle, so that they do not change with prediction steps. For example, if the equivalent flexible vibration frequency is calculated to be 10 Hz and the minimum constraint direction is a certain spatial vector in the current cycle, then all prediction steps within the entire prediction window will use this frequency and direction to avoid numerical instability of the prediction model due to frequent parameter changes.

[0077] The recursive derivation of coupled flexible vibration and rigid body motion includes: constructing discrete models for two types of states using a semi-implicit Euler method; considering the combined effects of the squared frequency term, damping term, and the projection of the driving acceleration in the recursive derivation of the flexible vibration state; and following the integral relationship between velocity and acceleration in the recursive derivation of the rigid body state. For example, in the modal velocity recursion, the modal velocity at the next moment is first calculated based on the current modal displacement, velocity, and driving acceleration, and then the modal displacement is updated with the new modal velocity to ensure numerical stability.

[0078] Timescale adaptation with a fixed number of steps includes scaling the system sampling period to the prediction sampling interval by adjusting the prediction step size coefficient. The larger the coefficient, the smaller the prediction sampling interval, and the longer the physical time covered by the fixed number of prediction steps. For example, with a system sampling period of 1 millisecond, an adjustment coefficient of 100, and a prediction sampling interval of 0.01 milliseconds, 20 prediction steps cover 0.2 milliseconds, which can completely capture the phase changes of slow modes.

[0079] The structural damping ratio constant is set as follows: 0.05 when the stiffness of the mechanism members is high; and 0.15 when the lightweighting of the members leads to significant vibration. The vibration decay time is verified through impulse response experiments to ensure that it is within a reasonable range.

[0080] Modal controlled gain constant identification includes: applying a pulse acceleration command with an amplitude of 0.5 m / s² along the minimum constraint direction to the end, measuring the end vibration response, and obtaining the actual value of the modal controlled gain constant by fitting a second-order system impulse response model.

[0081] The prediction step count includes 20 prediction steps. With a sampling period of 1 millisecond, the nominal operating condition covers a 20-millisecond prediction window, which can cover half a cycle of slow modes under most operating conditions, balancing computational efficiency and prediction completeness.

[0082] The modal state initialization method includes: using the modal displacement and modal velocity at the start of the prediction window, directly using the modal estimation results of the current control cycle (from the state estimation in claim 7), to ensure continuous connection between the prediction and the actual state.

[0083] The semi-implicit Euler stability conditions include: the prediction sampling interval should be less than one-tenth of the equivalent flexible vibration period. For example, if the equivalent flexible vibration frequency is 10 Hz and the period is 100 milliseconds, the prediction sampling interval should be less than 10 milliseconds to avoid numerical divergence.

[0084] Preferably, a multi-objective cost function is established, which includes weights based on the predicted step size adjustment coefficient. Driving acceleration commands are generated and executed through continuous numerical iteration. This includes: Step A, Constraint Construction: Use the Softplus function to construct constraints for the drive speed. With driving acceleration Continuous penalty items and : Step B, Cost Function: Establish the multi-objective cost function. It includes a trajectory error term, a platform energy consumption term, a modal energy term, a smoothing term, and the continuous penalty term: Step C, Weight Scaling: Adjusting the weights based on the predicted step size coefficients. Set the weight parameters in step B, where As a reference constant: Step D, Iterative Solution: Calculate the multi-objective cost function for the control sequence. gradient and execute a fixed number of times. Gradient descent update, outputting the first driving acceleration command. : Step E, Smooth Execution: Calculate the final issued drive joint velocity command using the hyperbolic tangent function. : In the formula, These are the driving speed and acceleration limits, respectively. For smoothing parameters; For prediction steps Positional error; The reflection inertia matrix of the platform; The modal displacements are those in the flexible vibration state of the structure. To constrain weights; This is the iteration step size; The damping least squares inverse matrix is ​​given. The system sampling period; This indicates element-wise multiplication.

[0085] The upper limit of drive speed is a parameter that limits the maximum rotational speed of the drive joint. It is preferably determined based on the rated parameters of the servo system to avoid damage to the mechanism due to overspeed of the joint.

[0086] The upper limit of drive acceleration is a parameter that limits the maximum acceleration of the end effector. It is preferably determined based on the structural strength of the mechanism and the servo output capability to balance gripping efficiency and structural safety.

[0087] The Softplus smoothing parameter is a parameter that adjusts the smoothness of continuous penalty terms. It is preferably between 20 and 80, so that the penalty terms in the non-exceeding area approach zero, and the exceeding area grows rapidly.

[0088] The constraint weight is a parameter that adjusts the intensity of the penalty for exceeding the speed and acceleration limits. It is preferably between 1 and 5 to avoid the penalty being too strong, which would lead to conservative control or too weak control.

[0089] The baseline weight is the basic weight constant of each component of the multi-objective cost function, preferably 1, to balance the priorities of each control objective in the initial state.

[0090] The iteration step size is the update step size of the gradient descent algorithm, preferably 0.05, to balance the optimization of convergence speed and numerical stability.

[0091] The fixed number of iterations is the number of iterations of the gradient descent algorithm, preferably 3, to balance real-time performance and optimization accuracy based on engineering experience.

[0092] The maximum gradient limit is a parameter that restricts the magnitude of the gradient of the cost function, preferably 5, to avoid gradient explosion leading to optimization divergence.

[0093] The position normalization scale is a parameter that unifies the dimensions of position error, preferably 0.01 meters, to meet the position accuracy requirements for grasping floating objects in the river.

[0094] The speed normalization scale is a parameter that unifies the dimensions of speed error, preferably 1 meter per second, to match the normal operating speed range of the mechanism.

[0095] The energy consumption normalization scale is a parameter that unifies the dimensions of energy consumption items, preferably 0.1 Nm, to calculate typical energy consumption values ​​based on the upper limit of drive-side inertia and acceleration.

[0096] The modal displacement normalization scale is a parameter that unifies the dimensions of modal displacement, preferably 0.001 meters, to represent the typical displacement range of structural vibration.

[0097] The modal velocity normalization scale is a parameter that unifies the dimensions of modal velocities, preferably 0.1 m / s, based on the typical velocity range of structural vibration.

[0098] The system sampling period is the time interval at which the control system repeatedly performs data processing.

[0099] The platform reflection inertia matrix is ​​an inertia matrix that reflects the end-effector dynamics.

[0100] Modal displacement is the displacement state of a structure's flexible vibration in the direction of minimum constraint.

[0101] The control sequence is a vector consisting of acceleration commands driven by each prediction step.

[0102] The gradient of the cost function is the vector of partial derivatives of the cost function with respect to the control sequence.

[0103] The drive acceleration command is the end acceleration control quantity output by the controller.

[0104] The command speed is the end velocity obtained by integrating the driving acceleration command.

[0105] The drive joint speed command is a speed command mapped to the drive side.

[0106] The speed command issued to drive the joint is the final speed command after smoothing and limiting.

[0107] Softplus constructs a continuously differentiable penalty term. This involves the Softplus function using logarithmic and exponential operations to make the penalty term approach zero in the non-limited region, without affecting normal control; and exhibiting non-linear growth in the limit-exceeding region, forming a strong constraint. For example, with a driving speed limit of 10 radians per second and a smoothing parameter of 50, when the actual speed is 11 radians per second, the penalty term increases significantly, forcing the optimization process to avoid exceeding the limit. This method avoids the non-smoothness of the optimization problem caused by hard constraints, ensuring the stability of the numerical solution.

[0108] The dynamic weighting strategy includes: the weights of the trajectory error term and energy consumption term decrease as the prediction step size adjustment coefficient increases, while the weights of the modal energy term and smoothing term increase as the adjustment coefficient increases. For example, when the adjustment coefficient is 10, the weight of the trajectory error term becomes 0.1 times the baseline value, and the weight of the modal energy term becomes 10 times the baseline value, realizing the near-singular region control target's transition from precise tracking to vibration suppression and smoothing.

[0109] The adjoint equation gradient calculation combined with fixed-iteration gradient descent includes: calculating the gradient recursively through the adjoint equation, eliminating the need to differentiate the control sequence one by one, significantly reducing computational cost; and using a fixed number of iterations to avoid branching logic in iterative convergence checks, ensuring real-time control. For example, a fixed number of three iterations, with each iteration updating the control sequence using the gradient, satisfies accuracy requirements while keeping computation time within the sampling period.

[0110] The hyperbolic tangent saturation combined with damped least squares inverse mapping includes: a hyperbolic tangent function to smoothly limit the command velocity, avoiding nonlinear jumps in the execution layer caused by discrete truncation; and a damped least squares inverse matrix to map the end-effector velocity to the drive joint space, balancing mapping accuracy and numerical stability in near-singular cases. For example, when the command velocity exceeds the limit, the hyperbolic tangent function smoothly compresses it to near the upper limit, preventing sudden velocity shocks to the mechanism.

[0111] The upper limit of drive speed and acceleration includes: the upper limit of drive speed is 80% of the rated speed of the servo motor. For example, if the rated speed of the servo is 15 radians per second, the upper limit of drive speed is set to 12 radians per second; the upper limit of drive acceleration is determined according to the strength test of the mechanism material. For example, for aluminum alloy rod mechanisms, it is 5 meters per second squared.

[0112] Softplus parameter adjustments include: setting the value to 80 when a stronger over-limit penalty is needed; setting the value to 20 when a gentler penalty is desired; and reducing the value to 30 and increasing the constraint weight to 5 if the value overflows.

[0113] The weight ratio relationship includes: the constraint weight is 3 and the baseline weight is 1 to ensure that the penalty for exceeding the limit is moderate; when the weight is dynamically scaled, the weights of the trajectory error term and the energy consumption term change synchronously, and the weights of the modal energy term and the smoothing term change synchronously, with the ratio remaining consistent.

[0114] The iteration step size includes: a fixed value of 0.05 is preferred, which can be increased to 0.1 if the optimization convergence is slow; if oscillation occurs, it can be reduced to 0.02, without the need for complex line search.

[0115] Control sequence initialization includes: setting all elements of the control sequence to 0 at startup; subsequent cycles use shift initialization, that is, shifting the control sequence of the previous cycle by one bit, while keeping the last bit unchanged.

[0116] The normalization scale includes: a position normalization scale of 0.01 meters, corresponding to the typical position error range of the grasping task; a velocity normalization scale of 1 meter per second, matching the normal operating speed of the mechanism; and an energy consumption normalization scale set at 0.1 Nm, calculated based on the product of the drive-side inertia and the upper limit of acceleration.

[0117] Preferably, the structural vibration energy at the end of the three-degree-of-freedom parallel mechanism in the direction of minimum constraint is monitored, and the model correction parameters used to calculate the critical response period are adaptively corrected based on the structural vibration energy, including: Step A, State estimation: Calculate the end-tracking error vector and project it onto the minimum constraint direction. Obtain the projection error scalar Then, first-order filtering is used to calculate and estimate the modal displacement. With estimated modal velocity : Step B, Energy Calculation: Using the equivalent flexible vibration frequency used to calculate the critical response period Calculate the vibration energy of the structure. : Step C, Parameter Update: Calculate the exponential mapping intermediate variable based on the structural vibration energy. Update volume And update after limiting using the hyperbolic tangent function. Finally, the model correction parameters for the next control cycle are calculated. : In the formula, This refers to the actual position of the end point; The final target location; These are the filter coefficients for modal displacement and modal velocity, respectively. The system sampling period; For adaptive gain updates; This is the single-cycle update limit constant; It is an exponential function with the natural constant as its base.

[0118] The actual position of the end effector is the real spatial position of the end effector of the three-degree-of-freedom parallel mechanism in the base coordinate system, which can be acquired by an encoder combined with forward kinematics or a vision sensor.

[0119] The end target position is the desired grab position of the floating object in the base coordinate system, which can be acquired by visual sensors or lidar.

[0120] The modal displacement filtering coefficient is a parameter that adjusts the smoothness of modal displacement estimation. It is preferably 0.0488, calculated based on a 20-millisecond filtering time constant, to balance response speed and noise resistance.

[0121] The modal velocity filter coefficient is a parameter that adjusts the smoothness of the modal velocity estimation. It is preferably 0.0952 and is calculated based on a 10-millisecond filter time constant to suppress differential noise.

[0122] The adaptive update gain is a parameter that controls the update speed of the model correction parameters. It is preferably 1e-3 to avoid system oscillation caused by excessively fast updates, while taking into account adaptive responsiveness.

[0123] The single-cycle update limit constant is a parameter that restricts the magnitude of a single update of the model correction parameters. It is preferably 0.05, with a maximum adjustment of 5% per cycle to ensure stable parameter updates.

[0124] The direction of minimum constraint is the spatial direction in which the transmission capability of the mechanism is weakest.

[0125] The equivalent flexible vibration frequency is a frequency parameter that reflects the slow-mode vibration characteristics of a structure.

[0126] The end-point tracking error vector is the difference between the actual position of the end point and the target position.

[0127] The projection error scalar is the projection value of the end-tracking error vector onto the direction of minimum constraint.

[0128] The estimated modal displacement is the structural flexible vibration displacement obtained through filtering.

[0129] The estimated modal velocities are obtained through filtering of the structural flexible vibration velocities.

[0130] Structural vibration energy is a physical quantity that characterizes the flexible modal strength of a structure.

[0131] The parameter update step size is the amount of time the model calibration parameters are updated in a single step.

[0132] The update step size after limiting is the parameter update amount after smoothing the limiting.

[0133] Exponential mapping intermediate variables are transitional variables used to ensure that the model calibration parameters are positive.

[0134] Model correction parameters are adjustment parameters that correct the equivalent flexible vibration frequency.

[0135] The end-point tracking error vector projection includes: taking the dot product of the end-point tracking error vector in 3D space with the minimum constraint direction vector, and extracting the error component along the weakest direction of the mechanism. For example, if the end-point tracking error vector is 0.03 m 0.01 m 0.02 m, and the minimum constraint direction vector is 0.8 0.1 0.1, then the projection error scalar is 0.03 × 0.8 + 0.01 × 0.1 + 0.02 × 0.1 = 0.027 m, focusing on the key error sources.

[0136] Modal state estimation includes: applying a first-order low-pass filter to the projection error scalar to obtain the estimated modal displacement, and then differentiating and filtering the modal displacement to obtain the estimated modal velocity. For example, if the projection error scalar is 0.027 meters, the modal displacement filter coefficient is 0.0488, and the estimated modal displacement in the previous cycle was 0.025 meters, then the current estimated modal displacement is (1-0.0488)×0.025+0.0488×0.027≈0.0251 meters, smoothing out noise interference.

[0137] Structural vibration energy construction includes: constructing an energy index based on the equivalent flexible vibration frequency, estimated modal displacement, and velocity, and the sum of kinetic and potential energy. For example, if the equivalent flexible vibration frequency is 10 Hz, the estimated modal displacement is 0.0251 m, and the estimated modal velocity is 0.01 m / s, then the vibration energy is 0.5 × (2π × 10⁻⁶). 2 ×0.0251 2 +0.5×0.01 2 ≈0.124 joules, quantifying vibration intensity.

[0138] Exponential mapping parameterization includes: representing the model correction parameters as exponential functions of the intermediate variables in the exponential mapping, ensuring that the parameters are always positive. For example, if the intermediate variable in the exponential mapping is 0.02, then the model correction parameter is the natural constant raised to the power of 0.02 ≈ 1.02, avoiding the distortion of the dynamic model caused by the parameter becoming negative.

[0139] Hyperbolic tangent limiting update includes: continuously limiting the parameter update step size. For example, if the update step size is 0.08 and the single-cycle update limiting constant is 0.05, then the update step size after limiting is 0.05 × hyperbolic tangent (0.08 / 0.05) = 0.05 × 0.989 ≈ 0.049 meters, to avoid excessively large single update amplitude.

[0140] The values ​​of the filtering coefficients include: the modal displacement filtering coefficient is calculated as the ratio of the negative sampling period (1 minus the natural constant) to the filtering time constant. For example, if the sampling period is 1 millisecond and the filtering time constant is 20 milliseconds, the coefficient is 1-exp(-0.001 / 0.02)≈0.0488; the modal velocity filtering coefficient is calculated similarly, and is 0.0952 when the filtering time constant is 10 milliseconds.

[0141] The adaptive update gain value range includes: a preferred value of 1e-3, which can be increased to 5e-3 if the vibration energy converges slowly, and decreased to 5e-4 if oscillation occurs, and adjusted according to the actual vibration suppression effect.

[0142] The single-cycle update limit constant is set to 0.05, which corresponds to a maximum change of 5% in the model correction parameters per cycle. This ensures the flexibility of adaptive correction while avoiding system instability caused by sudden parameter changes.

[0143] The initialization phase includes: when the system starts, the estimated modal displacement and velocity are initialized to 0. Iterative updates will begin after the projection error is acquired in the first control cycle to ensure the initial state is stable.

[0144] The initial values ​​of intermediate variables in the exponential mapping include: the initial value is set to 0, the initial value of the corresponding model correction parameter is the natural constant to the power of 0 = 1, the default dynamic model has no bias, and meets the initial control requirements.

[0145] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0146] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A method for controlling the trajectory of a floating object in a river channel for grasping objects with three degrees of freedom, characterized in that, include: The position and velocity data of the floating target are collected by sensors and fused with the current motion state data of the end effector of the three-degree-of-freedom parallel mechanism; The transmission singularity index characterizing the transmission capacity is calculated based on the geometric relationship of the links of the three-degree-of-freedom parallel mechanism, and the equivalent inertia at the end along the minimum constraint direction is obtained by combining the rotor inertia mapping of the drive motor. A critical response period reflecting the degree of hysteresis in the dynamic response of the mechanism is established based on the transmission singularity index and the end-effector equivalent inertia, and a predictive step size adjustment coefficient is generated based on the critical response period. The prediction sampling interval of the rolling planning controller is dynamically adjusted using the prediction step size adjustment coefficient, and a dynamic model containing the flexible vibration state of the structure is constructed within the rolling planning controller so that the prediction window covers the flexible vibration period under a fixed number of prediction steps. A multi-objective cost function is established, which includes weights based on the predicted step size adjustment coefficient. The driving acceleration command is generated and executed by continuous numerical iteration. The structural vibration energy at the end of the three-degree-of-freedom parallel mechanism in the direction of minimum constraint is monitored, and the model correction parameters used to calculate the critical response period are adaptively corrected based on the structural vibration energy.

2. The method for controlling the trajectory of a floating object in a river channel for grasping objects with three degrees of freedom as described in claim 1, characterized in that, The system collects and fuses the position and velocity data of the floating target with the current motion state data of the end effector of the three-degree-of-freedom parallel mechanism, including: Acquire timestamped sensor measurement data and use a recent sample-and-hold and first-order linear extrapolation strategy to perform time alignment processing on the sensor measurement data based on the time difference between the current control time and the measurement time; Using pre-calibrated rigid body transformation parameters, the time-aligned sensor measurement data is transformed from the sensor coordinate system to the target base coordinate system; A first-order exponential smoothing fusion algorithm is adopted. Using a fixed fusion coefficient determined by the sampling period and the filtering time constant, the historical state data of the previous control cycle and the sensor measurement data after conversion in the current control cycle are weighted and summed to obtain the fused position and velocity data of the floating target and the current motion state data of the end of the three-degree-of-freedom parallel mechanism.

3. The method for controlling the trajectory of a floating object in a river channel for grasping objects with three degrees of freedom as described in claim 1, characterized in that, The transmission singularity index, characterizing the transmission capacity, is calculated based on the geometric relationships of the links in the three-degree-of-freedom parallel mechanism. Combined with the rotor inertia mapping of the drive motor, the equivalent end inertia along the minimum constraint direction is obtained, including: A velocity Jacobian matrix is ​​constructed using the drive joint positions and link geometric parameters of the three-degree-of-freedom parallel mechanism. Singular value decomposition is then performed on the velocity Jacobian matrix to obtain the maximum singular value, the minimum singular value, and the characteristic direction vector corresponding to the minimum singular value. The ratio of the maximum singular value to the minimum singular value is used as the transmission singularity index, and the feature direction vector after continuous sign alignment is used as the minimum constraint direction. Construct a diagonal inertia matrix for the drive side that includes the inertia of the motor rotor, the inertia of the reducer, and the equivalent inertia of the connecting rod, and construct a damping least squares inverse matrix based on the speed Jacobian matrix and the adaptive damping coefficient; The diagonal inertia matrix of the drive side is mapped to the end platform side using the damping least squares inverse matrix to obtain the platform reflection inertia matrix, and the platform reflection inertia matrix is ​​projected onto the minimum constraint direction to obtain the end equivalent inertia.

4. The method for controlling the trajectory of a floating object in a river channel for grasping objects with three degrees of freedom as described in claim 3, characterized in that, A critical response period reflecting the degree of dynamic response hysteresis of the mechanism is established based on the transmission singularity index and the equivalent inertia of the end effector, and a predictive step size adjustment coefficient is generated based on the critical response period, including: A continuously varying surrogate stiffness is constructed using the minimum singular value obtained during the calculation of the transmission singularity index, wherein the surrogate stiffness decreases nonlinearly as the minimum singular value decreases. Divide the surrogate stiffness by the equivalent end inertia and take the square root to obtain the reference frequency. Multiply the reference frequency by the model correction parameter to obtain the equivalent flexible vibration frequency. The critical response period is obtained by calculating the ratio of pi to the equivalent flexible vibration frequency. Calculate the ratio of the critical response period to the preset reference time scale, and add the square of the ratio to the unit value to obtain the prediction step size adjustment coefficient.

5. The method for controlling the trajectory of a floating object in a river channel for grasping objects with three degrees of freedom as described in claim 4, characterized in that, The prediction sampling interval of the rolling planning controller is dynamically adjusted using the prediction step size adjustment coefficient, and a dynamic model including the flexible vibration state of the structure is constructed within the rolling planning controller to ensure that the prediction window covers the flexible vibration period under a fixed number of prediction steps, including: Divide the system sampling period by the prediction step size adjustment coefficient to obtain the prediction sampling interval, and lock the equivalent flexible vibration frequency and the minimum constraint direction as constants within the prediction window of the rolling planning controller; A discrete recursive relationship describing the evolution of the flexible vibration state of the structure is constructed using the predicted sampling interval, wherein the rate of change of the flexible vibration state of the structure is proportional to the square of the equivalent flexible vibration frequency and is affected by the damping term and the projection of the driving acceleration command onto the minimum constraint direction. The predicted sampling interval is used to construct a discrete recursive relationship describing the evolution of the rigid body's motion state, wherein the change in the rigid body's position is determined by the rigid body's velocity, and the change in the rigid body's velocity is determined by the driving acceleration command. The flexible vibration state of the structure is combined with the rigid motion state to form the predicted state variable of the rolling planning controller at each step within the prediction window.

6. The method for controlling the trajectory of a river floating object grasping object with three degrees of freedom as described in claim 5, characterized in that, A multi-objective cost function is established, which includes weights based on the predicted step size adjustment coefficient. Driving acceleration commands are generated and executed through continuous numerical iteration, including: A continuously differentiable penalty term is constructed using a smooth logarithmic exponential function to address the over-limit of the drive joint velocity and the over-limit of the drive acceleration. The continuously differentiable penalty term approaches zero in the non-over-limit region and exhibits nonlinear growth in the over-limit region. A multi-objective cost function is established by accumulating within the prediction window. The multi-objective cost function includes: a trajectory error term characterizing the end tracking accuracy, a platform energy consumption term characterizing the driving work, a modal energy term characterizing the amplitude of the flexible vibration state of the structure, a smoothing term characterizing the rate of change of the control quantity, and the continuously differentiable penalty term. The weights in the multi-objective cost function are dynamically set according to the prediction step size adjustment coefficient, wherein the weights of the trajectory error term and the platform energy consumption term are inversely proportional to the prediction step size adjustment coefficient, and the weights of the modal energy term and the smoothing term are directly proportional to the prediction step size adjustment coefficient. The gradient of the multi-objective cost function with respect to the control sequence is calculated using the adjoint equation, and a fixed number of gradient descent iterations are performed to update the current driving acceleration command. The command velocity obtained by integrating the drive acceleration command is smoothed and limited using a hyperbolic tangent saturation function, and the limited command is mapped to the drive joint space for execution.

7. The method for controlling the trajectory of a floating object in a river channel for grasping objects with three degrees of freedom as described in claim 1, characterized in that, Monitoring the structural vibration energy at the end of the three-degree-of-freedom parallel mechanism in the direction of minimum constraint, and adaptively correcting the model calibration parameters used to calculate the critical response period based on the structural vibration energy, including: The difference between the actual position and the target position of the end of the three-degree-of-freedom parallel mechanism is calculated to obtain the end tracking error vector, and the end tracking error vector is projected onto the minimum constraint direction to obtain the projection error scalar. The estimated modal displacement is obtained by performing a first-order low-pass filtering on the projection error scalar, and the estimated modal velocity is obtained by performing a first-order low-pass filtering on the differential rate of change of the estimated modal displacement. Using the equivalent flexible vibration frequency used to generate the critical response period, the estimated modal displacement, and the estimated modal velocity, the structural vibration energy characterizing the current structural flexible modal strength is constructed; Based on the gradient of the structural vibration energy with respect to the adaptive scaling factor, the parameter update step size for reducing the structural vibration energy is calculated, and the hyperbolic tangent function is used to continuously limit the parameter update step size. The step size of the updated parameters after the amplitude limit is used to update the intermediate variable of the exponential mapping, and the exponential function value of the intermediate variable of the exponential mapping is determined as the model correction parameter for the next control cycle.

Images