A mechanical arm predictive control method and system for unknown load

Abstract

The application discloses a kind of unknown load-oriented mechanical arm predictive control method and system, the method includes: grasping event detection and mode switching;Multiple sinusoidal limited detection excitation injection;Rolling time window response acquisition;Physical information network online update system matrix;Expansion state observer input compensation;With constraint MPC;Synthetic control input.The system includes: event determination module, excitation injection module, response acquisition module, matrix update module, input compensation module, constraint module and torque synthesis module.Through using the application, it can realize the closed-loop control of quick, stable and in accordance with base posture constraint after grasping.The application can be widely applied to the field of mechanical arm control.

Description

Technical Field

[0001] This invention relates to the field of robotic arm control, and more particularly to a predictive control method and system for robotic arms oriented towards unknown loads. Background Technology

[0002] In on-orbit servicing (OOS) missions, robotic arms often need to capture, grasp, and transport target satellites, debris, or payloads. After grasping, the system dynamics undergo abrupt changes due to uncertainties such as load mass, moment of inertia, grasping point, contact stiffness / damping, and clamping constraints, which couple with the base attitude / angular velocity to produce significant reaction effects. Under free-floating or weak attitude control conditions, the robotic arm and base form an underactuated coupled system. Joint torques will generate reaction effects on the base through momentum conservation, causing base attitude / angular velocity drift. On-orbit attitude control systems typically have limited bandwidth and actuator dead zones and saturation, making "maintaining stability and satisfying attitude constraints for a short period after grasping" a control challenge. If the controller still uses the pre-grab model or relies on insufficient online identification, "runaway" behaviors such as amplified joint oscillations, control saturation, cumulative increase in base angular velocity, and even instability are likely to occur.

[0003] The grasping process also introduces dynamic abrupt changes and high-frequency disturbances: the mass and moment of inertia of the unknown target, as well as the grasping point position, alter the system's equivalent inertia and coupling terms; the clamping contact introduces stiffness / damping changes; and collisions and friction cause the closed-loop poles and natural frequencies to drift. If the pre-grab model or slow-converging online identification and updating is still used, it is easy to cause the accumulation of prediction biases, misjudgment of constraints, and amplification of oscillations. In addition, on-orbit missions usually require stable handling, docking, or manipulation as soon as possible after grasping, leaving a very short time window for "remodeling," which places higher demands on real-time performance and safety.

[0004] Most existing methods are based on the calculation of torque, robust / adaptive control, sliding mode control, etc., which are based on accurate dynamic models. These methods perform well when the model is well matched, but they are sensitive to sudden changes in the model after capture and it is difficult to obtain a usable predictive model in a short time, thus reducing the control accuracy. Summary of the Invention

[0005] In view of this, in order to solve the technical problem that most existing robotic arm control methods require accurate models, and it is difficult to obtain a usable accurate model in a short period of time after a grasping event occurs, thus leading to low control accuracy, this invention proposes a predictive control method for robotic arms oriented towards unknown loads. This method includes the following steps: Once the robotic arm has completed its grasping action, the system immediately switches the control mode from normal trajectory tracking or free operation mode to a rapid stabilization control mode after grasping. Based on the nominal control torque, a multi-sine probe torque with limited amplitude is superimposed to excite the robotic arm-base system. During the excitation application period, the state response sequence of the robotic arm and base within a rolling time window is collected. The collected state response sequence is input into the physical information network to update the system matrix online, outputting a discrete state space matrix. Using this discrete state space matrix and an extended state observer, the input compensation is calculated. A prediction model is constructed using the discrete state space matrix, and the nominal control torque is output under preset constraints. Finally, the nominal control torque, input compensation, and probe torque are combined to form the final control torque.

[0006] In addition to the above method, the present invention also proposes a predictive control system for a robotic arm with unknown loads. The system includes: an event determination module, an excitation injection module, a response acquisition module, a matrix update module, an input compensation module, a constraint module, and a torque synthesis module.

[0007] Based on the above scheme, this invention provides a predictive control method and system for a robotic arm oriented towards unknown loads. By superimposing controllable small-amplitude multi-sine probes into the control torque and collecting the system response sequence within a short window after grasping, it directly maps the system response sequence into a discrete state-space system matrix that can be used for predictive control. This avoids the long-term excitation and iterative parameter updates required by traditional online identification. At the same time, it uses an extended state observer (ESO) to compensate for residual errors in the form of equivalent inputs, so that even if the prediction model still has deviations in the early stage, it can maintain closed-loop stability and gradually improve prediction accuracy, thereby achieving fast, stable closed-loop control that conforms to the base attitude constraints after grasping. Attached Figure Description

[0008] Figure 1 This is a flowchart of the steps of a predictive control method for a robotic arm oriented to unknown loads according to the present invention. Figure 2 This is a structural block diagram of a predictive control system for a robotic arm oriented to unknown loads according to the present invention. Detailed Implementation

[0009] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0010] It should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0011] It should be understood that the terms "system," "apparatus," "unit," and / or "module" used in this application are a method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0012] In the description of the embodiments of this application, "a plurality of" refers to two or more. The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0013] Furthermore, flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, the steps can be processed in reverse order or simultaneously. Additionally, other operations can be added to these processes, or one or more steps can be removed from them.

[0014] Reference Figure 1 The diagram below illustrates an optional example of the predictive control method for a robotic arm under unknown loads proposed in this invention. This method can be applied to computer devices, and the predictive control method proposed in this embodiment may include, but is not limited to, the following steps: Step S1: A grabbing event is detected, and the control mode is switched; Step S2: Based on the nominal control torque, superimpose a multi-sine detection torque with limited amplitude to perform excitation injection; Step S3: During the excitation injection period, collect the state response sequence of the robot arm and base within the rolling time window; Step S4: Input the state response sequence into the physical information network and generate a discrete state space matrix; Step S5: Generate input compensation based on the discrete state space matrix and the extended state observer; Step S6: Construct a prediction model based on the discrete state space matrix and output the nominal control torque in combination with preset constraints; Step S7: Generate the final control torque based on the nominal control torque, input compensation, and probe torque.

[0015] An integrated process of "grasping event triggering – detection excitation – rapid matrix update – observation compensation – constraint predictive control" is proposed for an on-orbit robotic arm after grasping an unknown load. This process achieves stable, real-time closed-loop control that meets the base attitude constraints for a short period of time after grasping by using event-driven, short-term excitation, rapid update of the system matrix by physical information network, compensation of residual error by extended state observer and constraint MPC closed-loop control.

[0016] Nominal control torque: refers to the basic control torque used to drive the robotic arm to perform tasks, calculated by solving the MPC optimization problem based on the current state.

[0017] The nominal control torque output in step S6 is the updated value and is used as the input data for step S7.

[0018] In some feasible embodiments, step S1 specifically includes: During the robotic arm's grasping process, multiple signals, including gripper current, sudden changes in motor torque, end-effector force / torque thresholds, and grasping duration, are monitored. When at least one criterion is met, the robotic arm is deemed to have completed the grasping action. The system immediately switches the control mode from normal trajectory tracking or free operation mode to a rapid stabilization control mode after grasping. This mode switch ensures timely initiation of subsequent rapid modeling and closed-loop control processes, while preventing excessive disturbances during grasping transients from causing base attitude drift or joint oscillations. The mode switch can be configured with hysteresis or delay filtering to prevent false triggering, and single-stage or multi-stage triggering strategies can be selected according to task requirements.

[0019] In some feasible embodiments, it also includes: Discretize the robotic arm system, and define a system state vector for each control cycle: in Joint angle, The joint angular velocity, The base angular velocity, This represents the number of joints in the robotic arm. The system input is the joint torque. The control objective is to calculate a closed-loop torque based on the current state that can suppress oscillations, deviation accumulation, and base attitude disturbances caused by grasping unknown loads.

[0020] In some feasible embodiments, step S2, the expression for the multi-sine detection torque, is as follows: in , and These represent the amplitude, frequency, and initial phase of each sinusoidal component, and can be scheduled based on joint torque saturation, joint velocity / position limiting, and base angular velocity disturbance constraints.

[0021] By selecting different frequency components to cover the system's inherent frequency range, the excitation sequence can quickly reflect the system dynamic changes caused by unknown loads without causing excessive disturbances to the actuators or base. Furthermore, a multi-sine sequence superposition strategy can be set according to task requirements, including sequential excitation, alternating excitation, or random phase superposition, to improve the robustness and generalization ability of the system matrix estimation.

[0022] In some feasible embodiments, step S3, the state response sequence, is represented as follows: in Joint angle, The joint angular velocity, The base angular velocity, This represents the total torque actually applied. Optional data acquisition of joint acceleration is also available. Alternatively, end-effector pose information can be used to enhance network input features. The collected data undergoes normalization, filtering for noise reduction, and time alignment to form the input tensor of the physical information network. This is used for online updates of the system matrix. The length N of the rolling time window can be adjusted according to the control cycle and the dynamic characteristics of the task to ensure that it can capture short-term system dynamics after capture without exceeding the stability window allowed in orbit.

[0023] In some embodiments, the rolling time window can also be constructed by collecting state-input data pairs from the most recent N sampling periods: After denoising, normalization, and time alignment, the input tensor of the physical information network is formed. The choice of the scroll window length N should take into account: sufficient to capture dynamic abrupt changes and coupling effects after grabbing; and not exceeding the stability control window allowed by the on-orbit mission.

[0024] Optional extensions: for Increase joint acceleration Or end-effector pose / velocity information, to improve the network's ability to perceive nonlinear coupling and higher-order dynamics.

[0025] In some feasible embodiments, step S4 involves inputting the scroll window. Inputting a pre-trained physical information network yields a discrete state space matrix, which is represented as follows: Optional output disturbance channel Output matrix Used to extend the prediction model. The matrix can be output directly. Incremental output And iterate and update: During the offline training phase, the network employs domain randomization to cover load mass, inertia, gripping point variations, and contact stiffness / damping. Discrete dynamic residuals, bounded energy, and closed-loop stability constraints are also incorporated to ensure the physical feasibility of the online inference matrix.

[0026] By updating via a rolling time window, the system can quickly obtain an effective matrix that can be used for predictive control within a short period of time after the capture is completed, avoiding the long-term excitation and iterative convergence required by traditional identification methods.

[0027] In some feasible embodiments, step S5 specifically includes: The deviation between the network output matrix and the actual system, the dynamic mismatch caused by unknown loads, and external / coupling disturbances are combined into an extended state. : ESO online estimation is obtained : Generate equivalent input compensation based on the estimation results: Or the inverse form of regularization in This is the regularization coefficient, ensuring stability when the B matrix is ​​not invertible or has a condition number difference.

[0028] The compensation input acts directly on the control channel, which can offset the impact of modeling errors and unknown disturbances on the closed-loop stability in real time, and improve the system's robustness to handling unknown loads.

[0029] In some feasible embodiments, step S6 specifically includes: based on Build a prediction model in the prediction time domain Minimize the weighted quadratic cost function within the inner region: in, This represents the objective function to be optimized. Indicates the prediction time domain, Indicates the first The predicted state vector of the step, Indicates the first The reference state vector of the step. Indicates the index of prediction steps. Represents the state weight matrix. Indicates the first Step control increment, This represents the control increment weight matrix. Indicates the first The base perturbation estimation state of the step, This represents the base perturbation weight matrix. This represents the minimum value of the joint angle. This represents the maximum value of the joint angle. Indicates the first The joint angle of the step, This represents the minimum value of the joint angular velocity. This represents the maximum value of the joint angular velocity. Indicates the first The joint angular velocity of the step, This represents the minimum value of the control torque. This indicates the maximum value of the control torque. Indicates the first Step control torque, This represents the lower limit of the constraint on the base disturbance state. This represents the upper limit of the constraint on the base disturbance state.

[0030] By combining explicit constraints such as joint position / velocity / torque limits, base angular velocity constraints, and end-effector task constraints, design a constraint MPC: Joint position Joint velocity Joint torque Angular velocity of the base .

[0031] The MPC solution employs strategies such as hot start, fixed number of iterations, or reduced-order prediction models to ensure optimization is completed within the control cycle. The output is the nominal control torque. .

[0032] MPC enables trajectory tracking, disturbance suppression, and closed-loop performance optimization while meeting constraints.

[0033] In some feasible embodiments, step S7, the final control torque synthesis is: Or The control torque is uniformly allocated within the MPC constraints. It is applied to the joint actuators to achieve rapid and stable control after grasping, suppressing runaway behavior and meeting base attitude constraints and task execution requirements within a short timeframe. Through this integrated process, system matrix reconstruction, residual compensation, and closed-loop control can be completed within a short time window after grasping an unknown load, enabling rapid switching between stable handling, docking, or operational tasks after grasping.

[0034] Based on the overall process described above, the beneficial effects of this invention also include: short-window response sequence → PINN rolling update of the system matrix, reducing the convergence time of traditional identification iteration; ESO observation network output error and unknown load mismatch and input compensation, reducing the risk of instability; MPC explicit processing of constraints such as joints / bases, reducing the probability of constraint violation; network inference and MPC adopt fixed computation budget / warm start, adapting to on-orbit computing power; and it can be integrated into existing control systems as a dedicated control mode after grasping.

[0035] like Figure 2 As shown, a predictive control system for a robotic arm accommodating unknown loads includes: The event determination module is used to execute step S1; The stimulus injection module is used to execute step S2; The response acquisition module is used to execute step S3; The matrix update module is used to execute step S4; The input compensation module is used to execute step S5; The constraint module is used to execute step S6; The torque synthesis module is used to execute step S7.

[0036] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0037] A predictive control device for robotic arms with unknown loads: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a predictive control method for a robotic arm oriented to an unknown load as described above.

[0038] The content of the above method embodiments is applicable to the device embodiments. The specific functions implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0039] A storage medium storing processor-executable instructions, which, when executed by a processor, are used to implement a predictive control method for a robotic arm oriented to unknown loads as described above.

[0040] The content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0041] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A predictive control method for a robotic arm accommodating unknown loads, characterized in that, Includes the following steps: Based on the nominal control torque, a multi-sine detection torque with limited amplitude is superimposed for excitation injection; During the excitation injection period, the state response sequence of the robotic arm and base within the rolling time window is collected; The state response sequence is input into the physical information network to generate a discrete state space matrix; Based on the discrete state space matrix and the extended state observer, input compensation is generated; A prediction model is constructed based on the discrete state space matrix, and a nominal control torque is output in combination with preset constraints; The final control torque is generated based on the nominal control torque, the input compensation, and the detection torque.

2. The predictive control method for a robotic arm oriented towards unknown loads as described in claim 1 is characterized in that, The step of superimposing a multi-sine detection torque with limited amplitude on the nominal control torque for excitation injection further includes: A grabbing event was detected; control mode was switched.

3. The predictive control method for a robotic arm oriented towards unknown loads according to claim 2, characterized in that, The expression for the multi-sine detection torque is as follows: in, express The constantly injected detection torque, Represents a time variable. This represents the total number of superimposed sinusoidal components. Indicates the first A sine component , Indicates the first The amplitude of each component, Indicates the first The frequency of each component Indicates the first The initial phase of each component.

4. The predictive control method for a robotic arm oriented towards unknown loads according to claim 2, characterized in that, The state response sequence is represented as follows: in, Joint angle, The joint angular velocity, The base angular velocity, This represents the total torque actually applied.

5. The predictive control method for a robotic arm oriented towards unknown loads according to claim 4, characterized in that, The step of generating input compensation based on the discrete state space matrix and the extended state observer specifically includes: The discrete state space matrix, system deviation, dynamic mismatch caused by unknown load, and coupling disturbance are combined into an extended state; The expansion state and the expansion state observer are combined to estimate and generate input compensation.

6. The predictive control method for a robotic arm oriented towards unknown loads according to claim 5, characterized in that, The expression for the input compensation is as follows: in, Indicates input compensation. This represents the estimation result of the extended state observer.

7. The predictive control method for a robotic arm oriented towards unknown loads according to claim 1, characterized in that, The step of constructing a prediction model based on the discrete state space matrix and outputting the nominal control torque in conjunction with preset constraints is expressed by the following formula: in, This represents the objective function to be optimized. Indicates the prediction time domain, Indicates the first The predicted state vector of the step, Indicates the first The reference state vector of the step. Indicates the index of prediction steps. Represents the state weight matrix. Indicates the first Step control increment, This represents the control increment weight matrix. Indicates the first The base perturbation estimation state of the step, This represents the base perturbation weight matrix. This represents the minimum value of the joint angle. This represents the maximum value of the joint angle. Indicates the first The joint angle of the step, This represents the minimum value of the joint angular velocity. This represents the maximum value of the joint angular velocity. Indicates the first The joint angular velocity of the step, This represents the minimum value of the control torque. This indicates the maximum value of the control torque. Indicates the first Step control torque, This represents the lower limit of the constraint on the base disturbance state. This represents the upper limit of the constraint on the base disturbance state.

8. A predictive control system for a robotic arm accommodating unknown loads, characterized in that, include: The excitation injection module is used to superimpose a multi-sine probe torque with limited amplitude on the nominal control torque for excitation injection. The response acquisition module is used to acquire the state response sequence of the robotic arm and base within a rolling time window during the excitation injection period; The matrix update module inputs the state response sequence into the physical information network and generates a discrete state space matrix. The input compensation module generates input compensation based on the discrete state space matrix and the extended state observer; The constraint module constructs a prediction model based on the discrete state space matrix and outputs a nominal control torque in combination with preset constraints. The torque synthesis module generates the final control torque based on the nominal control torque, the input compensation, and the detected torque.

9. A predictive control device for a robotic arm accommodating unknown loads, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a predictive control method for a robotic arm oriented to an unknown load as described in any one of claims 1-7.

Images