Multi-scene application intelligent robot and multi-scene control method thereof

Abstract

The present application belongs to the general control or regulation system technical field, and relates to a kind of multi-scene application intelligent robot and its multi-scene control method, comprising: obtaining the motion characteristic vector under the operation scene of power response component;The motion characteristic vector is projected to control law space to determine inertia bias estimate value;In scene switching transient, according to the associated response of the initial control weight matrix of target scene and inertia bias estimate value, mapping transient gain partial derivative;Determine whether transient gain partial derivative exceeds output stability threshold, and then modulate initial control weight matrix according to the change rate of inertia bias estimate value, generate evolution control instruction that meets frequency response limit, the present application constructs the prior constraint mechanism of physical inertia, integrates dynamic momentum evolution trend into control logic evolution process, eliminates the physical impact of discrete instruction on continuous mechanical entity, ensures that the transfer function of switching transient evolves smoothly, enhances the dynamic robustness of regulation system.

Description

Technical Field

[0001] This invention belongs to the field of general control or regulation system technology, and in particular relates to a multi-scenario application intelligent robot and its multi-scenario control method. Background Technology

[0002] Current conventional control or regulation systems mainly calibrate parameters for preset stable operating conditions to ensure the steady-state accuracy of the controlled object under specific environments.

[0003] With the increasing demands for flexible operations, controlled objects need to switch between heterogeneous workspaces with different dynamic characteristics. Due to the physical mismatch between the discrete update cycle of the control logic and the mechanical inertial response capability of the actuator, the system experiences a step evolution of the control law at the moment of work condition switching, resulting in energy jumps within the closed-loop system. Physically, this manifests as overshoot, oscillation, or physical shock in the end effector. For example, Chinese invention patent application CN105643627A discloses a gain adjustment device and method for robot motion control. The technical solution detects motor current and speed, identifies load inertia using the least squares method, and adjusts the gain parameters accordingly. Analysis of the mechanism reveals that this solution involves adjusting the gain parameters after sampling steady-state data. Post-adjustment relies on the completion of load switching and the system entering a relatively stable operating phase. During the millisecond-level transient process of multi-scenario switching, there is a lack of prior prediction of the evolution trend of physical momentum. There is a physical mismatch between the discrete jumps of control parameters and the continuous evolution of mechanical inertia. Energy jumps cannot be avoided within the closed-loop system, manifesting as overshoot, oscillation, or physical shocks in the end-drive components. The industry generally adopts the method of switching multiple sets of preset parameters to deal with this. However, during cross-scenario transients, the system's state variables are difficult to maintain continuity at the switching point. Conventional technical approaches include increasing the number of filter stages or extending the control transition time. However, increasing the number of filter stages generates additional phase lag, reducing the system's response bandwidth, while extending the transition time weakens the production efficiency of process connections.

[0004] Therefore, how to achieve synchronous evolution of control logic commands and physical inertial states, and establish closed-loop safety constraint boundaries that can resist model mismatch and environmental disturbances, has become the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention aims to solve the problem of sudden jumps in system energy and a decrease in steady-state accuracy caused by the decoupling of discrete control commands and continuous physical inertia during scene switching.

[0006] In this technical solution, a multi-scenario application intelligent robot and its multi-scenario control method are provided, which, for the controller, includes the following steps: Step S1: Obtain the motion feature vector of the dynamic response component in the current working scenario. The motion feature vector consists of real-time angular velocity and instantaneous acceleration to characterize the dynamic physical momentum of the dynamic response component. Step S2: Project the motion feature vector onto the preset control law space, and obtain the inertial bias estimate reflecting the physical inertial trend of the dynamic response component by calculating the control damping coefficient of the dynamic response component in the current motion state. Step S3: During the transient process of the intelligent robot switching the work scene, the initial control weight matrix corresponding to the target scene is obtained, and the nonlinear correlation response between the inertial bias estimate and the initial control weight matrix is ​​calculated using the dynamic amplitude limiter, and the transient gain partial derivative under the switching transient is obtained. Step S4: Determine whether the transient gain partial derivative exceeds the preset output stability threshold of the dynamic response component. The output stability threshold is jointly calibrated based on the rated torque limit of the dynamic response component and the current saturation constraint of the servo driver. Step S5: If the transient gain partial derivative exceeds the output stability threshold, the initial control weight matrix is ​​modulated by amplitude attenuation based on the rate of change of the inertial bias estimate in the time domain to generate an evolution control command limited by the frequency response limit of the dynamic response component, so that the transfer function of the dynamic response component in the transient process of scene switching evolves smoothly along the energy envelope of the control law space.

[0007] Preferably, in step S1, the original feedback parameters of the dynamic response components are collected by the inertial measurement unit and encoder built into the robot body, and the original feedback parameters are identified online by the recursive least squares algorithm to remove random environmental noise and extract the real-time angular velocity and instantaneous acceleration.

[0008] Preferably, in step S2, the Jacobian matrix, which reflects the evolution trend of the control law space, is invoked to map the motion feature vector to the null space of the Jacobian matrix, and the control damping coefficient of the dynamic response component in the current motion state in response to the change of the control law is calculated to obtain the estimated value of the inertial bias.

[0009] Preferably, in step S3, the norm change rate of the initial control weight matrix is ​​calculated, and the norm change rate is multiplied by the inertial bias estimate to obtain the dynamic load variation of the dynamic response component under the control command step, and the dynamic load variation is defined as the transient gain partial derivative.

[0010] Preferably, after generating the evolution control command, the state deviation of the dynamic response component in the target scenario is monitored in real time. A parameter compensation operator is constructed based on the real-time modulus of the inertial bias estimate to offset the environmental disturbance torque. The evolution control command is corrected through the parameter compensation operator so that the state deviation conforms to the preset attenuation envelope.

[0011] Preferably, before step S1, a candidate control law database covering different working conditions is established, multimodal sensing parameters of the current environment are obtained, and the corresponding control law space is matched from the candidate control law database according to the multimodal sensing parameters. The multimodal sensing parameters include ambient light characteristics, spatial position coordinates, and contact torque characteristics.

[0012] Preferably, the construction process of the parameter compensation operator includes: obtaining the environmental contact torque characteristics at the end of the dynamic response component, calculating the cross-correlation coefficient between the environmental contact torque characteristics and the inertial bias estimate, adjusting the gain weight of the parameter compensation operator based on the cross-correlation coefficient, and suppressing signal oscillations caused by unstructured environmental interference.

[0013] Preferably, the output stability threshold defines the safety envelope boundary of the intelligent robot when operating across scenarios. It dynamically calibrates the maximum allowable gain of the power response component in the transient process by calculating the mechanical load limit of the power response component in real time, combined with the thermal loss constraint of the servo driver and the maximum torque output limit.

[0014] Preferably, in step S5, the amplitude change rate of the evolution control command in the time domain is not higher than 50Hz, and the second derivative of the evolution control command is limited by the mechanical resonant frequency boundary of the dynamic response component, so as to avoid the forced vibration of the dynamic response component induced by the switching of the control weight matrix and to ensure the mechanical life of the controlled system.

[0015] Compared with existing technologies, the multi-scenario application intelligent robot and its multi-scenario control method of this invention have the following advantages: 1. In intelligent robots used in multiple scenarios, by projecting the motion feature vector of the execution unit onto the control law vector space, the generation of control commands is given a priori constraints of physical inertia. This mechanism does not simply rely on error feedback to perform lag adjustment, but starts from the evolution trend of the underlying physical momentum and pre-modulates the evolution rate of the control topology, thereby eliminating the hard impact of discrete logic commands on continuous mechanical entities and ensuring the smooth evolution of the transfer function of the controlled object in the transient process of cross-scenario switching.

[0016] 2. Due to the introduction of the energy envelope constraint, the system can evaluate the energy transition gradient generated by the switching between the old and new control weights in real time. When the gradient exceeds the stability threshold of the mechanical structure, the controller maintains the physical balance of the control gain by dynamically scaling the weight matrix. This mechanism realizes the real-time synchronization of the response capability of the control logic and the execution unit, avoiding overshoot, oscillation or physical collision caused by model mismatch in the unstructured environment, and effectively enhancing the dynamic stability of the regulation system.

[0017] 3. By combining the local regulation law velocimetry and the rate of change of the predicted projection vector, a nonlinear damping regulation loop is constructed. The loop uses the motion characteristics of the execution unit itself to guide the correction of the control parameters in reverse, so that the steady-state accuracy of the system no longer depends on the reconstruction of the full model in a higher dimension. Instead, it uses lightweight weight offsetting to offset the disturbance torque caused by the sudden change in environmental impedance. This multi-level control mechanism coupling not only ensures the industrial-grade real-time response speed, but also ensures the control accuracy and convergence level of the regulation system under dynamic loads. Attached Figure Description

[0018] Figure 1 This is a flowchart of the intelligent robot multi-scenario smooth evolution control method of the present invention; Figure 2 This is a diagram of the architecture of the multimodal sensing and adaptive parameter compensation control system of this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] It should be noted that all directional and positional terms used in this invention, such as: up, down, left, right, front, back, vertical, horizontal, inner, outer, top, bottom, transverse, longitudinal, center, etc., are only used to explain the relative positional relationship and connection between components in a specific state (as shown in the accompanying drawings). They are only for the convenience of describing this invention and do not require that this invention be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the descriptions of "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0021] In the description of this invention, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0022] In the description of this specification, references to the terms "an embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example, and the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0023] A multi-scenario intelligent robot and its multi-scenario control method, used in a controller, includes the following steps: Step S1: Obtain the motion feature vector of the dynamic response component in the current working scenario. The motion feature vector consists of real-time angular velocity and instantaneous acceleration to characterize the dynamic physical momentum of the dynamic response component. Step S2: Project the motion feature vector onto the preset control law space, and obtain the inertial bias estimate reflecting the physical inertial trend of the dynamic response component by calculating the control damping coefficient of the dynamic response component in the current motion state. Step S3: During the transient process of the intelligent robot switching the work scene, the initial control weight matrix corresponding to the target scene is obtained, and the nonlinear correlation response between the inertial bias estimate and the initial control weight matrix is ​​calculated using the dynamic amplitude limiter, and the transient gain partial derivative under the switching transient is obtained. Step S4: Determine whether the transient gain partial derivative exceeds the preset output stability threshold of the dynamic response component. The output stability threshold is jointly calibrated based on the rated torque limit of the dynamic response component and the current saturation constraint of the servo driver. Step S5: If the transient gain partial derivative exceeds the output stability threshold, the initial control weight matrix is ​​modulated by amplitude attenuation based on the rate of change of the inertial bias estimate in the time domain to generate an evolution control command limited by the frequency response limit of the dynamic response component, so that the transfer function of the dynamic response component in the transient process of scene switching evolves smoothly along the energy envelope of the control law space.

[0024] Preferably, in step S1, the original feedback parameters of the dynamic response components are collected by the inertial measurement unit and encoder built into the robot body, and the original feedback parameters are identified online by the recursive least squares algorithm to remove random environmental noise and extract the real-time angular velocity and instantaneous acceleration.

[0025] Preferably, in step S2, the Jacobian matrix, which reflects the evolution trend of the control law space, is invoked to map the motion feature vector to the null space of the Jacobian matrix, and the control damping coefficient of the dynamic response component in the current motion state in response to the change of the control law is calculated to obtain the estimated value of the inertial bias.

[0026] Preferably, in step S3, the norm change rate of the initial control weight matrix is ​​calculated, and the norm change rate is multiplied by the inertial bias estimate to obtain the dynamic load variation of the dynamic response component under the control command step, and the dynamic load variation is defined as the transient gain partial derivative.

[0027] Preferably, after generating the evolution control command, the state deviation of the dynamic response component in the target scenario is monitored in real time. A parameter compensation operator is constructed based on the real-time modulus of the inertial bias estimate to offset the environmental disturbance torque. The evolution control command is corrected through the parameter compensation operator so that the state deviation conforms to the preset attenuation envelope.

[0028] Preferably, before step S1, a candidate control law database covering different working conditions is established, multimodal sensing parameters of the current environment are obtained, and the corresponding control law space is matched from the candidate control law database according to the multimodal sensing parameters. The multimodal sensing parameters include ambient light characteristics, spatial position coordinates, and contact torque characteristics.

[0029] Preferably, the construction process of the parameter compensation operator includes: obtaining the environmental contact torque characteristics at the end of the dynamic response component, calculating the cross-correlation coefficient between the environmental contact torque characteristics and the inertial bias estimate, adjusting the gain weight of the parameter compensation operator based on the cross-correlation coefficient, and suppressing signal oscillations caused by unstructured environmental interference.

[0030] Preferably, the output stability threshold defines the safety envelope boundary of the intelligent robot when operating across scenarios. It dynamically calibrates the maximum allowable gain of the power response component in the transient process by calculating the mechanical load limit of the power response component in real time, combined with the thermal loss constraint of the servo driver and the maximum torque output limit.

[0031] Preferably, in step S5, the amplitude change rate of the evolution control command in the time domain is not higher than 50Hz, and the second derivative of the evolution control command is limited by the mechanical resonant frequency boundary of the dynamic response component, so as to avoid the forced vibration of the dynamic response component induced by the switching of the control weight matrix and to ensure the mechanical life of the controlled system.

[0032] Example 1: In an unmanned matrix factory application scenario involving parallel heterogeneous tasks, the intelligent robot needs to transition from milling operations with high rigidity interaction characteristics to precision assembly operations with high flexibility position control requirements. Due to physical fluctuations in the dynamic characteristics of the controlled objects between work scenarios, the transfer function of the system undergoes a step change during the transient process of scenario switching. Under the control path, the control unit loads the initial control weight matrix of the target scenario at the moment of switching. This discrete logical instruction update and the momentum state of the dynamic response components with physical inertia cause a mismatch, which induces energy jumps within the closed-loop system. At the physical level, this manifests as overshoot, oscillation, or physical impact of the end effector. This not only limits the residual convergence speed of the system after cross-scenario switching, but also causes mechanical wear on the actuator due to instantaneous torque fluctuations, constituting a technical bottleneck that limits the dynamic robustness of the control system.

[0033] When the system detects a trigger signal indicating a switch from milling to assembly operations, the control unit acquires the motion characteristic vector of the power response component in real time under the current operating scenario. This motion characteristic vector, composed of real-time angular velocity and instantaneous acceleration acquired by the built-in inertial measurement unit, characterizes the dynamic physical momentum of the power response component. The control unit projects the motion characteristic vector onto a preset control law space. By invoking the Jacobian matrix, which reflects the evolution trend of the control law space, the control unit maps the motion characteristic vector to the null space of the Jacobian matrix and calculates the control damping coefficient of the power response component in response to changes in the control law under the current motion state. This yields an estimated inertial bias value reflecting the physical inertial trend of the power response component. Based on the statistical laws of discrete-time signals, the parameter identification algorithm is ensured to converge rapidly within a short transient switching window of 50ms. The control unit has a built-in dynamic reset procedure for the covariance matrix and monitors the control logic transition edge. When the initial control weight matrix of the target scenario is issued, the initial covariance matrix of the recursive least squares algorithm is directly replaced with the historical covariance convergence final value obtained in advance under the corresponding steady-state conditions. The forgetting factor is set in the range of 0.90 to 0.95. The determination of this range comes from the experimental results of the step dynamic response of the ultimate load. When the applied forgetting factor is lower than 0.90, the experimental monitoring shows that the algorithm forgets the historical covariance too much, causing the parameter iteration path to diverge, which in turn induces the observation residual to exceed the control threshold significantly. When this value is higher than 0.95, the proportion of retained historical memory causes overshoot in the estimated convergence steps, resulting in significant time tailing in the output and failing to meet the rigid timing constraint requirement of a 50ms transient switching window. Therefore, bilateral limits are calibrated to determine the optimal attenuation working region. Inheriting the noise statistical characteristics of the historical steady-state data system, the algorithm reduces the number of iterations under stimulated transients, outputting high signal-to-noise ratio acceleration estimates within an extremely short time window. This eliminates the hysteresis error caused by spatiotemporal scale mismatch from the source of physical measurement. Based on the robot's rigid body dynamics model, the true evolution of physical momentum depends on the end-effector's pose constraints and the reconstruction of the joint space's internal state. The mapping process uses a generalized Jacobian matrix weighted pseudo-inverse decomposition path, based on the formula... The generalized Jacobian matrix (GBM) with inertial features is calculated. The motion eigenvectors are orthogonally projected onto the GBM value space and null space, respectively. The value space projection component represents the torque feedforward necessary to maintain the continuity of the end-effector trajectory, while the null space projection component corresponds to the release of redundant momentum within the end-effector pose without altering its position. The null space projection component is extracted and calibrated as the control damping coefficient. The generalized Jacobian matrix represents the bridge between momentum and physical mapping of the operation space. The joint space real-time mass inertia matrix represents the current true mass distribution characteristics of the system. and These represent the fundamental Jacobian matrix and its transpose derived from geometric constraints, respectively. The autoencoder reads real-time joint angular displacements. The introduction of the mechanical matrix establishes the objective basis and solution path for the transformation from pure kinematic state vectors to dynamic damping coefficients. After obtaining the initial control weight matrix corresponding to the target scenario, the control unit uses a dynamic amplitude limiter to calculate the estimated inertial bias value. The nonlinear correlation response between the initial control weight matrix and the transient gain partial derivative under switching transients is mapped to obtain the transient gain partial derivative, and it is determined whether the transient gain partial derivative exceeds the preset output stability threshold of the dynamic response component. The output stability threshold is jointly calibrated based on the rated torque limit of the dynamic response component and the current saturation constraint of the servo driver to define the safety envelope boundary of the intelligent robot when operating across scenarios. According to the law of conservation of electromechanical energy conversion, the gain change rate of the logic control domain is objectively mapped to the output torque pulsation of the physical execution domain, unifying the internal logic judgment dimension and the underlying physical torque dimension of the system. The control unit uses the preset system comprehensive impedance coefficient to obtain the transient gain partial derivative. Implement dimensional transformation; the physical dimension transformation correlation is as follows: The expected value of the equivalent moment fluctuation is obtained by solving. ,in, The representative has the following characteristics after conversion: Expected value of the fluctuation of the equivalent moment of an absolute physical quantity; The representative impedance coefficient, which characterizes the current electromagnetic and mechanical coupling dynamic stiffness of the controlled system, is established by extracting the peak amplitude-frequency response at the resonance peak through a preliminary no-load frequency sweep experiment of the driving system. The expected value of the equivalent torque fluctuation will be obtained in the subsequent comparison and judgment step. dimensionless Output stability threshold Numerical comparison is used to eliminate logical gaps in cross-dimensional parameter comparisons. If the transient gain partial derivative exceeds the output stability threshold, the control unit estimates the value based on the inertial bias. The rate of change in the time domain is modulated with amplitude attenuation on the initial control weight matrix to generate an evolutionary control command constrained by the frequency response limit of the dynamic response component. This allows the transfer function of the dynamic response component to evolve smoothly along the energy envelope of the control law space during scene switching transients. The generation of the evolutionary control command follows the following logic: ;in, Evolutionary control commands are output to the driver. The basic topology weights for the target scene. The corrected weights are generated by an adaptive algorithm. The inertial sensitivity coefficient, The inertial bias estimate is obtained by introducing the rate of change of the predicted projection vector. This system achieves nonlinear damping adjustment of the control gain, ensuring that the evolution rate of the control command is spatiotemporally synchronized with the inertial response capability of the physical entity. During the calculation of the amplitude attenuation control logic, the system pre-extracts the Euclidean norm of the multidimensional rate of change of the predicted projection vector. The extracted norm scalar value is used as the actual characteristic parameter substituted into the rate of change term in the denominator of the equation for arithmetic operations. This legally reduces the high-dimensional vector features to the scalar domain, ensuring that the mathematical dimensions of each term in the denominator of the logical expression are consistent and that there is no physical directional conflict. This ensures that the underlying damping adjustment algorithm follows the linear algebra matrix operation criteria, transforming the evolved control command generated by amplitude attenuation modulation into a forced constraint action of the underlying hardware to execute the feedforward torque limiting procedure, and extracting the evolved control command. The transient fundamental amplitude is converted into a target current command signal adapted to the servo drive. A first-order low-pass digital filter is used to extract the high-frequency step component in the smooth current command signal. The smoothed current signal is written into the inverter module's pulse width modulation duty cycle control register via a digital-to-analog converter channel. The duty cycle of the insulated gate bipolar transistor is adjusted to limit the output from the underlying electrical execution circuit to the instantaneous excitation current of the motor stator winding, thus constructing a physical output safety boundary. By constructing the above-mentioned prior constraint mechanism based on physical inertia, the system eliminates the physical impact of discrete commands on continuous mechanical entities, reducing the torque fluctuation rate of the intelligent robot during the switching between milling and assembly scenarios to less than 5%, and improving the residual convergence speed by more than 40%. Under the condition of maintaining a 1000Hz update frequency, the control logic command and the physical inertial state evolve at the same frequency, ensuring the steady-state accuracy and mechanical life of the adjustment system under unstructured environmental interference.

[0034] Example 2: In a typical control or regulation system performance verification test, the test platform includes a dynamic response component with a six-degree-of-freedom motion envelope, a control unit with a signal acquisition interface, and a built-in inertial measurement unit integrated at the end of the dynamic response component. The built-in inertial measurement unit has an angular velocity measurement resolution of 0.005 degrees per second and an acceleration sampling accuracy of 0.01g, used to capture high-frequency momentum fluctuations during scene switching transients in real time. The test data originates from the feedback stream of physical sensors. Gaussian white noise with a signal-to-noise ratio of 20dB and industrial electromagnetic environment harmonics with a frequency of 50Hz are superimposed on the acquired signal to simulate interference in an industrial production environment. Regarding the core parameter settings, the sampling period... The determination is based on the balance between the spectral bandwidth of the monitored signal and the processor load. When the characteristic frequency of the dynamic response component is in the 100Hz frequency range, in order to satisfy the Nyquist sampling law and the processor load is less than 60%, [the following parameters are used]. Set to 1ms.

[0035] The experimental design establishes a multidimensional control system, including an experimental group using the method of this invention, a control group A with the energy envelope verification mechanism removed, and a control group with the inertial sensitivity coefficient... Control group B, with a load range outside 0.5 to 1.5, under gradient verification during the transition from a 5kg milling scenario to a 1kg precision assembly scenario, showed an overshoot of 0.12mm at the end of the transition transient, a torque fluctuation rate of 4.2%, and a residual convergence time of 112.4ms. Control group A, under the same transition conditions, exhibited an overshoot of 1.85mm, a torque fluctuation rate increasing to 22.6%, and a current step impact at the driver. Control group B, however, showed an overshoot of 1.85mm under the same transition conditions. When the value is set to 2.5, the dynamic response speed of the system decreases, the residual convergence time is extended to 358.6ms, and oscillations with a frequency of 12.5Hz occur. This indicates that the parameter range defined in this invention is the working window for achieving a trade-off between control stability and response speed. As the intensity gradient of scene switching increases, i.e., the load fluctuation increases from 2kg to 10kg, the transient gain partial derivative of the test group output shows a monotonically increasing trend. The control unit adjusts the amplitude of the evolution control command to limit the energy transition of the system to the rated torque limit of the dynamic response component. Measurement data shows that under the highest intensity switching condition, the transmission efficiency of the test group remains at 94.5%, and the overcurrent protection action of the servo drive is not triggered. The above test process confirms that by constructing a priori constraint mechanism based on physical inertia, the physical impact of discrete commands on continuous mechanical entities is eliminated, thus meeting the steady-state adjustment requirements of general control or regulation systems in multi-scenario applications.

[0036] Example 3: This example combines Figures 1 to 2 This document describes a multi-scenario intelligent robot and its multi-scenario control method, such as... Figure 1As shown, step S1 obtains the motion characteristic vector of the dynamic response component under the current working scenario, which is composed of real-time angular velocity and instantaneous acceleration, to characterize dynamic physical momentum. Then, in step S2, the motion characteristic vector is projected onto the control law space. By calculating the control damping coefficient of the dynamic response component under the current motion state, the inertial bias estimate reflecting the physical inertial trend is obtained. In step S3, the initial control weight matrix of the target scenario is obtained in the transient state of switching the working scenario. The nonlinear correlation response between the initial control weight matrix and the inertial bias estimate is calculated using a dynamic amplitude limiter to map the transient gain partial derivative. In step S4, it is determined whether the transient gain partial derivative exceeds the output stability threshold jointly calibrated based on the rated torque limit of the dynamic response component and the current saturation constraint of the servo driver. Finally, in step S5, if it is determined that the threshold is exceeded, the amplitude attenuation modulation is performed on the initial control weight matrix according to the time-domain change rate of the inertial bias estimate to generate an evolution control command limited by the frequency response limit, so that the transfer function evolves smoothly along the energy envelope.

[0037] like Figure 2 As shown, the ambient lighting features, serving as the visual perception input stream, and the spatial position coordinates, serving as the geometric motion feedback stream, are synchronized in a multimodal data stream. This is combined with a candidate control law database for multimodal parameter scene matching to achieve scene semantic feature recognition, thereby matching the corresponding control law space. Simultaneously, the dynamic physical momentum trend is extracted through the inertial bias estimation core and input along with the contact torque features, serving as the physical contact force feedback stream, into the cross-correlation coefficient calculation stage. This correlation is used to link the inertial bias and contact torque, thus entering the parameter compensation operator construction process to build a nonlinear damping adjustment loop. Finally, through evolutionary control... To suppress unstructured environmental signal oscillations, in the specific operation of constructing a nonlinear damping adjustment loop, the system maps and transforms the cross-correlation coefficient through a set exponential smoothing activation function. When the absolute value of the cross-correlation coefficient approaches the safety threshold, it indicates that the environmental interference and physical momentum are strongly coupled at high frequency. The activation function immediately outputs a nonlinearly decaying weight value to reduce the gain ratio of the parameter compensation operator. Conversely, if the absolute value of the cross-correlation coefficient is within the safety threshold, a constant bias is output to maintain the original gain weight, thereby realizing the accurate and continuous transformation of the physical correlation signal into the compensation parameters inside the controller.

[0038] Example 4: In the application scenario of assembling high-precision micro-mechanical parts, the intelligent robot needs to switch between grasping conditions with dynamic response characteristics and assembly conditions with steady-state accuracy requirements. The physical size of the micro-mechanical parts is on the micrometer scale. The torque jump in the transient process of scene switching causes the energy jump of the power response component. If the jump exceeds the current saturation constraint of the servo driver, it will cause the motor winding to overheat or the control closed loop to fail.

[0039] Before operation, the system determines the output stability threshold using the following calibration procedure. The control unit reads the rated torque of the power response component. and the maximum undistorted output current of the servo driver And establish a linear mapping relationship between current and torque. ;,in, For torque, It is the torque constant. The control unit drives the power response component at rated speed under no-load conditions, using current, and measures and records the inherent mechanical friction loss of the system and the torque residual caused by the transmission backlash. Furthermore, according to the formula The output stability threshold is calculated. ;in, As a safety reserve factor, its value ranges from 0.85 to 0.95, used to compensate for motor resistance drift caused by ambient temperature fluctuations. When the system detects a trigger signal indicating a switch from the gripping mode to the assembly mode, the dynamic amplitude limiter inside the control unit calculates the transient gain partial derivative. The control unit extracts the inertial bias estimate at the current sampling time. Initial control weight matrix of the target scene Secondly, the control unit calculates... exist The projection norm in the operator space yields an intermediate variable characterizing the effect of physical momentum on the intensity of topological impact. The control unit uses a second-order Taylor expansion to calculate intermediate variables. The instantaneous rate of change in the time domain, and this rate of change is defined as the transient gain partial derivative. To establish the convergence benchmark of this expansion and maintain characteristic homology with the product of the underlying norm change rate, the system uses the time anchor point where the scene switching capture moment is located as the coordinate base point of the Taylor expansion. This allows the system to obtain a high-order approximation solution of the norm change component of the initial weight matrix in the zero neighborhood. The system then performs a feedback calculation by multiplying this instantaneous rate of change scalar solution in the time domain with the modulus of the current periodic inertial bias estimate. This achieves a comprehensive evolution path for reconstructing the amplitude of transient dynamic load changes through scalar product. If the determination... Greater than the output stability threshold This indicates that the update rate of the discrete instructions exceeds the response bandwidth of the mechanical entity, and the control unit initiates amplitude attenuation modulation.

[0040] The control unit estimates the value based on the inertial bias. The rate of change of the control command output to the servo axis is nonlinearly damped to ensure that the slope of the control command envelope evolution meets the physical tracking limit of the dynamic response component. Under the operating conditions of an ambient temperature of 25℃ and a sampling frequency of 1000Hz, the output stability threshold determined by the above calibration procedure reduces the current saturation rate of the system by 98.6%, and the torque ripple of the end effector is controlled within 3.5% of the rated value within a 0.05s scene switching time. By converting physical limits into logical constraints, the mismatch between control commands and mechanical inertia is eliminated, and steady-state regulation of the adjustment system in micron-level assembly tasks is achieved.

[0041] Example 5: In the commissioning of an automated production line involving the replacement of the end effector of the dynamic response component, due to the different mass distribution and link geometry characteristics of the new actuator compared to the preset model, the Jacobian matrix and its null space projection operator of the system deviate from the current physical inertial state. This causes the estimated inertial bias calculated by the control unit to deviate during the transient process of switching from the gripping operation to the precision assembly operation. The presence of uncalibrated eccentric torque disturbances induces an overshoot of more than 0.5 mm at the end of the dynamic response component, which hinders the smooth loading of the control topology weight matrix under zero error conditions and constitutes a bottleneck that limits the consistency of dynamic performance of general control or regulation systems after hardware changes.

[0042] When the control unit detects the trigger command for the initial hardware state change, the drive dynamic response component executes a set of low-speed frequency sweep trajectories covering the entire working envelope. Simultaneously, it extracts the joint encoder readings and the angular velocity and acceleration data output by the built-in inertial measurement unit through the signal acquisition interface. It then uses the singular value decomposition algorithm to reconstruct the basis vector coordinate system in the control law space, redetermines the transfer gain relationship between the joint space velocity and the motion characteristic vector, and obtains the corrected Jacobian matrix, thereby eliminating the influence of the end effector mass distribution change on the calculation of the control damping coefficient.

[0043] Through the above frequency sweep calibration process, the control unit will adjust the inertial sensitivity coefficient. The convergence residual was reduced to below 0.01%, and a dynamic consistency model of the dynamic response component under assembly conditions was established. The measured data showed that after completing three sets of frequency sweep cycles, the system achieved stable control of torque fluctuation rate within a 0.05s scene switching time, enabling the dynamic response component to maintain a command update frequency of 1000Hz after the actuator was replaced, thus meeting the physical stability requirements of the regulation system in a non-ideal deployment environment.

[0044] Example 6: In a system initialization scenario involving the synchronization of control commands for dynamic response components, the control unit determines the system's adjustment period based on the communication link delay and the intrinsic frequency of the mechanical entity. The system extracts the structural vibration period of the dynamic response components. And the processing time required for the control unit to calculate the initial control weight matrix once. And in accordance with the judgment criteria Lock in the adjustment cycle under the current operating conditions; among which, For the structural vibration period, The time consumed by logical operations To determine the adjustment period, when the structural vibration frequency of the dynamic response component is 500Hz and the time for a single logic operation is 0.8ms, the system-locked adjustment period meets the physical requirements for capturing the high-frequency inertial changes of the mechanical entity.

[0045] When the system faces the reconstruction of the Jacobian matrix and its null space operator, the control unit determines the orthogonal basis vectors of the control law space by decomposing the acquired motion feature vectors. Specifically, it calculates the distribution gradient of the motion feature vectors in the generalized space and selects the singular value component carrying more than 99% of the physical momentum information as the spatial projection principal axis, thereby eliminating the modeling residuals caused by the drift of the end load center of gravity. On this basis, the system drives the dynamic response component to carry out three sets of controlled frequency sweep trajectories within the full working envelope and monitors the inertial sensitivity coefficient in real time. The evolution trend continues until the coefficient deviation of two consecutive cycles is less than 0.005%, achieving synchronization between the physical inertial state and the evolution logic of the control command.

[0046] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A multi-scenario application intelligent robot and its multi-scenario control method, used in a controller, characterized in that, Includes the following steps: Step S1: Obtain the motion feature vector of the dynamic response component in the current working scenario. The motion feature vector consists of real-time angular velocity and instantaneous acceleration to characterize the dynamic physical momentum of the dynamic response component. Step S2: Project the motion feature vector onto the preset control law space, and obtain the inertial bias estimate reflecting the physical inertial trend of the dynamic response component by calculating the control damping coefficient of the dynamic response component in the current motion state. Step S3: During the transient process of the intelligent robot switching the work scene, the initial control weight matrix corresponding to the target scene is obtained, and the nonlinear correlation response between the inertial bias estimate and the initial control weight matrix is ​​calculated using the dynamic amplitude limiter, and the transient gain partial derivative under the switching transient is obtained. Step S4: Determine whether the transient gain partial derivative exceeds the preset output stability threshold of the dynamic response component. The output stability threshold is jointly calibrated based on the rated torque limit of the dynamic response component and the current saturation constraint of the servo driver. Step S5: If the transient gain partial derivative exceeds the output stability threshold, the initial control weight matrix is ​​modulated by amplitude attenuation based on the rate of change of the inertial bias estimate in the time domain to generate an evolution control command limited by the frequency response limit of the dynamic response component, so that the transfer function of the dynamic response component in the transient process of scene switching evolves smoothly along the energy envelope of the control law space.

2. The intelligent robot for multi-scenario applications and its multi-scenario control method according to claim 1, characterized in that, In step S1, the original feedback parameters of the dynamic response components are collected by the inertial measurement unit and encoder built into the robot body. The recursive least squares algorithm is used to identify the original feedback parameters online, remove random environmental noise, and extract the real-time angular velocity and instantaneous acceleration.

3. The intelligent robot for multi-scenario applications and its multi-scenario control method according to claim 1, characterized in that, In step S2, the Jacobian matrix, which reflects the evolution trend of the control law space, is invoked to map the motion feature vector to the null space of the Jacobian matrix. The control damping coefficient of the dynamic response component in the current motion state in response to the change of the control law is calculated to obtain the estimated value of the inertial bias.

4. The intelligent robot for multi-scenario applications and its multi-scenario control method according to claim 1, characterized in that, In step S3, the norm change rate of the initial control weight matrix is ​​calculated, and the norm change rate is multiplied by the inertial bias estimate to obtain the dynamic load variation of the dynamic response component under the control command step, and the dynamic load variation is defined as the transient gain partial derivative.

5. The intelligent robot for multi-scenario applications and its multi-scenario control method according to claim 1, characterized in that, After generating the evolution control command, the state deviation of the dynamic response component in the target scenario is monitored in real time. Based on the real-time modulus of the inertial bias estimate, a parameter compensation operator is constructed to offset the environmental disturbance torque. The evolution control command is corrected through the parameter compensation operator so that the state deviation conforms to the preset attenuation envelope.

6. The intelligent robot for multi-scenario applications and its multi-scenario control method according to claim 1, characterized in that, Before step S1, a candidate control law database covering different working conditions is established, multimodal sensing parameters of the current environment are obtained, and the corresponding control law space is matched from the candidate control law database according to the multimodal sensing parameters. The multimodal sensing parameters include ambient light characteristics, spatial position coordinates, and contact torque characteristics.

7. The intelligent robot for multi-scenario applications and its multi-scenario control method according to claim 5, characterized in that, The construction process of the parameter compensation operator includes: obtaining the environmental contact torque characteristics at the end of the dynamic response component, calculating the cross-correlation coefficient between the environmental contact torque characteristics and the inertial bias estimate, adjusting the gain weight of the parameter compensation operator based on the cross-correlation coefficient, and suppressing signal oscillations caused by unstructured environmental interference.

8. The intelligent robot for multi-scenario applications and its multi-scenario control method according to claim 1, characterized in that, The output stability threshold defines the safety envelope boundary of the intelligent robot when operating across different scenarios. It dynamically calibrates the maximum allowable gain of the power response component in the transient process by calculating the mechanical load limit of the power response component in real time, combined with the thermal loss constraint of the servo driver and the maximum torque output limit.

9. The intelligent robot for multi-scenario applications and its multi-scenario control method according to claim 1, characterized in that, In step S5, the amplitude change rate of the evolution control command in the time domain is no higher than 50Hz, and the second derivative of the evolution control command is limited by the mechanical resonant frequency boundary of the dynamic response component, so as to avoid the forced vibration of the dynamic response component induced by the switching of the control weight matrix and to ensure the mechanical life of the controlled system.

Images