Data-driven control methods and systems for scalable variable-stiffness continuum manipulators

Abstract

A data-driven control method and system for a scalable variable stiffness continuous body manipulator are disclosed. The data-driven control method includes: during the current prediction period, using the current state data and multiple sets of motion data of the scalable variable stiffness continuous body manipulator, based on a trained dynamic forward model, predicting multiple sets of state data for the next time step; dynamically adjusting multiple weights corresponding to multiple terms included in a comprehensive loss function based on task stage and tactile data; determining a set of state data that minimizes the comprehensive loss function after weight adjustment; and during the current control period, driving the scalable variable stiffness continuous body manipulator to execute actions corresponding to the determined set of motion data based on the set of motion data corresponding to the determined set of state data. Therefore, the data-driven control method can control a scalable variable stiffness continuous body manipulator exhibiting strong nonlinearity, time-varying, hysteresis, and strong correlation with the current state dynamic characteristics.

Description

Technical Field

[0001] This disclosure relates to the field of continuum robot control technology, and more specifically, to a data-driven control method and system for a telescopic variable stiffness continuum manipulator, particularly suitable for complex dynamic environments requiring flexible and safe interaction, such as animal husbandry, home services, medical care, and industry. Background Technology

[0002] Continuous robotic arms have wide potential applications in confined or non-cooperative environments due to their high degrees of freedom, inherent pneumatic compliance, and intrinsic safety. In the prior art, as disclosed in the applicant's previous application "A Stretchable Variable Stiffness Continuous Robotic Arm" (application number 202511820126.8), extension, bending, and stiffness adjustment are achieved through the antagonism between pneumatic pressure and a pull wire, and a tactile sensor is integrated, providing passive safety capabilities.

[0003] However, firstly, the compressibility of gas in this type of robotic arm results in highly nonlinear, time-varying, and hysteretic dynamic characteristics that are strongly correlated with the current state, making it difficult for traditional control methods to achieve high-precision, multi-objective (pose, extension, stiffness, and velocity) coordinated control. Secondly, existing data-driven soft robot control methods cannot adapt to complex task scenarios requiring online stiffness adjustment, velocity control, and active obstacle avoidance. Thirdly, existing methods often use random actions or simple frequency sweeps for data acquisition, leading to low data efficiency and poor model generalization. Summary of the Invention

[0004] The purpose of this disclosure is to provide a data-driven control method and system for controlling a scalable variable stiffness continuum manipulator with dynamic characteristics that exhibit strong nonlinearity, time-varying, hysteresis, and strong correlation with the current state.

[0005] According to one aspect of the embodiments of this disclosure, a data-driven control method for a scalable variable stiffness continuous body manipulator is provided. The data-driven control method includes: during a current prediction period, using the current state data and multiple sets of motion data of the scalable variable stiffness continuous body manipulator at the current moment, based on a trained dynamic forward model, predicting multiple sets of state data of the scalable variable stiffness continuous body manipulator at the next moment corresponding to the multiple sets of motion data; during the current prediction period, dynamically adjusting multiple weights corresponding to multiple terms included in a comprehensive loss function based on the task stage of the scalable variable stiffness continuous body manipulator and tactile data from a tactile sensor; during the current prediction period, determining a set of state data from the multiple sets of state data at the next moment that minimizes the comprehensive loss function after weight adjustment; and during a current control period corresponding to the current prediction period, driving the scalable variable stiffness continuous body manipulator to perform actions corresponding to the determined set of motion data based on the set of motion data corresponding to the determined set of state data.

[0006] Optionally, the data-driven control method further includes training a dynamic forward model. The step of training the dynamic forward model includes using dynamic response data acquired by applying at least one of a sweep frequency excitation signal, a step excitation signal, and a random combination excitation signal to a scalable variable stiffness continuum manipulator as preliminary training data to perform preliminary training on the dynamic forward model.

[0007] Optionally, the data-driven control method further includes: during the current control period, collecting air pressure regulation data and the corresponding response data of the extendable variable stiffness continuum manipulator as historical air pressure-response data, wherein the step of training the dynamic forward model further includes: periodically using the historical air pressure-response data to perform additional training on the dynamic forward model.

[0008] Optionally, during the current prediction period, the step of dynamically adjusting multiple weights corresponding to multiple terms included in the comprehensive loss function based on the task phase of the scalable variable stiffness continuum manipulator and the tactile data from the tactile sensor includes: increasing the weight of the term associated with the position of the scalable variable stiffness continuum manipulator among the multiple terms included in the comprehensive loss function when the task phase corresponds to the operation phase in which the scalable variable stiffness continuum manipulator performs an operation on the target; increasing the weight of the term associated with the stiffness of the scalable variable stiffness continuum manipulator among the multiple terms included in the comprehensive loss function when the task phase corresponds to the operation phase in which the scalable variable stiffness continuum manipulator performs an operation on the target; and increasing the weight of the term associated with obstacle avoidance penalty among the multiple terms included in the comprehensive loss function when a collision is detected by tactile data.

[0009] Optionally, the data-driven control method further includes: in response to tactile data exceeding a safety threshold, stopping the stretchable variable stiffness continuous body manipulator from executing the action corresponding to the set of action data, and driving the stretchable variable stiffness continuous body manipulator to execute a retraction action. The dynamic forward model is a temporal neural network with a memory unit capable of processing multi-dimensional temporal data. The dynamic forward model employs a multi-encoder fusion structure capable of handling heterogeneous inputs from position, cable position and force, manipulator posture, stiffness, air pressure, and tactile feedback. The step of determining the set of state data that minimizes the weighted comprehensive loss function from the multiple sets of state data at the next moment is implemented through a learnable policy network. The dynamic forward model is used as an environment simulator, taking the current state of the stretchable variable stiffness continuous body manipulator and the target task as input, and the negative value of the comprehensive loss function as a reward signal. The policy network is trained through offline or online reinforcement learning to output the corresponding action data.

[0010] Optionally, the input data of the dynamic forward model includes state data, motion data, and module position labels. The state data includes at least one of barometric pressure sensor and tactile sensor data. The motion data includes at least one of cable actuation and barometric pressure regulation. The module position labels are used to characterize the relative order of each segment of the telescopic variable stiffness continuum manipulator in the continuum structure, so that the dynamic forward model can distinguish the individual dynamic characteristics of different segments. The input data includes at least one of multiple terms in the comprehensive loss function, including end-effector pose loss, telescopic loss, stiffness loss, velocity loss, obstacle avoidance penalty, and motion smoothness penalty. The current control period and the next prediction period at least partially overlap.

[0011] According to one aspect of the embodiments of this disclosure, a data-driven control system for a scalable variable stiffness continuous body manipulator is provided. The data-driven control system includes: a prediction module configured to: during a current prediction period, using the current state data and multiple sets of motion data of the scalable variable stiffness continuous body manipulator at the current moment, based on a trained dynamic forward model, predict multiple sets of state data of the scalable variable stiffness continuous body manipulator at the next moment, corresponding to the multiple sets of motion data respectively; a weight adjustment module configured to: during the current prediction period, dynamically adjust multiple weights corresponding to multiple terms included in a comprehensive loss function based on the task stage of the scalable variable stiffness continuous body manipulator and tactile data from a tactile sensor; a determination module configured to: during the current prediction period, determine a set of state data from the multiple sets of state data at the next moment that minimizes the comprehensive loss function after weight adjustment from the multiple sets of state data; and an execution module configured to: during a current control period corresponding to the current prediction period, drive the scalable variable stiffness continuous body manipulator to execute actions corresponding to the determined set of motion data based on the determined set of motion data.

[0012] According to one aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions stored in the computer-readable storage medium are executed by at least one processor, the at least one processor causes the at least one processor to perform the data-driven control method described above for a telescopic variable stiffness continuous body robotic arm.

[0013] According to one aspect of an embodiment of the present disclosure, a computer program product is provided. The computer program product includes computer instructions that, when executed by at least one processor, implement the data-driven control method described above for a telescopic variable stiffness continuous body robotic arm.

[0014] The data-driven control method and system for a scalable variable stiffness continuum manipulator according to embodiments of the present disclosure can control a scalable variable stiffness continuum manipulator with dynamic characteristics exhibiting strong nonlinearity, time-varying, hysteresis, and strong correlation with the current state. This enables high-precision, multi-objective (pose, extension, stiffness, velocity) collaborative control of the scalable variable stiffness continuum manipulator, adapts to complex task scenarios requiring online stiffness adjustment, velocity control, and active obstacle avoidance, and improves data efficiency and model generalization. Attached Figure Description

[0015] The above and / or other aspects of this disclosure will become clearer and more readily understood from the following detailed description taken in conjunction with the accompanying drawings.

[0016] Figure 1 This is a flowchart illustrating a data-driven control method for a telescopic variable stiffness continuous body robotic arm according to an embodiment of the present disclosure.

[0017] Figure 2 This is a flowchart illustrating a method for training a dynamic forward model according to an embodiment of the present disclosure.

[0018] Figure 3 This is a block diagram illustrating a data-driven control system for a telescopic variable stiffness continuous body robotic arm according to an embodiment of the present disclosure.

[0019] Throughout the accompanying drawings and detailed embodiments, unless otherwise described or provided, the same reference numerals will be understood to denote the same elements, features, and structures. The drawings may not be to scale, and for clarity, illustration, and convenience, the relative dimensions, scale, and depiction of elements in the drawings may be exaggerated. Detailed Implementation

[0020] The following detailed embodiments are provided to aid the reader in gaining a comprehensive understanding of the methods, apparatus, and / or systems described herein. However, various changes, modifications, and equivalents of the methods, apparatus, and / or systems described herein will become apparent upon understanding this disclosure. For example, the order of operations described herein is merely illustrative and is not limited to those orders set forth herein, but can be appropriately altered upon understanding this disclosure, except for operations that must occur in a specific order. Furthermore, for clarity and brevity, descriptions of features known upon understanding this disclosure may be omitted.

[0021] The features described herein may be implemented in different forms and should not be construed as being limited to the examples described herein. Rather, the examples described herein are provided merely to illustrate some of the many feasible ways of implementing the methods, apparatus, and / or systems described herein that will be clear upon understanding the disclosure of this application.

[0022] Although terms such as “first,” “second,” and “third” may be used herein to describe various components, assemblies, regions, layers, or parts, these components, assemblies, regions, layers, or parts are not limited by these terms. Rather, these terms are used only to distinguish one component, assembly, region, layer, or part from another. Therefore, without departing from the teaching of the examples described herein, the first component, first assembly, first region, first layer, or first part referred to as the first component, first assembly, first region, first layer, or first part may also be referred to as the second component, second assembly, second region, second layer, or second part.

[0023] The terminology used herein is for the purpose of describing various examples only and is not intended to limit disclosure. Unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. The terms “comprising,” “including,” and “having” indicate the presence of the features, quantities, operations, components, elements, and / or combinations thereof stated therein, but do not preclude the presence or addition of one or more other features, quantities, operations, components, elements, and / or combinations thereof.

[0024] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as understood based on the disclosure of this application and as commonly understood by one of ordinary skill in the art to which this disclosure pertains. Unless expressly defined herein, terms (such as those defined in a general dictionary) shall be interpreted as having the same meaning as they have in the context of the relevant art and in the disclosure of this application, and shall not be interpreted ideally or overly formally. The use of the term “may” herein with respect to examples or embodiments (e.g., regarding what an example or embodiment may include or implement) indicates the existence of at least one example or embodiment that includes or implements such a feature, while not all examples are limited thereto.

[0025] Unless otherwise specified, the same reference numerals generally refer to the same elements (e.g., components, steps, and methods). Reference numerals described in previous embodiments that reappear in later embodiments may be omitted. Furthermore, technical features described in different or the same embodiments can be combined in any way, as long as the combined embodiment or technical solution is complete and can solve the technical problems of this application or achieve the technical effects described or not described in this disclosure but which can be determined based on the complete technical solution described above. The terminology used in this disclosure is explained below.

[0026] It should be noted that, where there is no conflict between the various embodiments, these embodiments and their features can be combined with each other.

[0027] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the system described above can be found in the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0028] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0029] Figure 1 This is a flowchart illustrating a data-driven control method for a telescopic variable stiffness continuous body robotic arm according to an embodiment of the present disclosure.

[0030] Reference Figure 1 The data-driven control method for a telescopic variable stiffness continuous body robotic arm according to embodiments of the present disclosure includes steps S100 to S400.

[0031] In step S100, during the current prediction period, using the current state data and multiple sets of motion data of the scalable variable stiffness continuous body manipulator, based on the trained dynamic forward model, the next set of state data of the scalable variable stiffness continuous body manipulator corresponding to the multiple sets of motion data is predicted.

[0032] According to embodiments of this disclosure, a telescopic variable stiffness continuous robotic arm (hereinafter referred to as the "robotic arm") may have the same structure as the robotic arm in the patent "A Telescopic Variable Stiffness Continuous Robotic Arm" (application number 202511820126.8), which may include end caps, covers, and a continuous structure composed of one or more continuous segments connected in series. Each continuous segment includes an airtight and watertight skin structure, interface assemblies located at both ends of the skin, clamping assemblies, a pull-wire drive device deployed inside the segment, and a pneumatic pressure regulating chamber. The robotic arm may integrate a tactile sensor (wound around the outside of the skin) and multiple pneumatic pressure sensors (distributed within each pneumatic pressure chamber). By adjusting the antagonism between the pneumatic pressure and the pull-wire tension, the extension, bending, and stiffness changes of the robotic arm can be achieved.

[0033] According to embodiments of this disclosure, a dynamic forward model can take the current state data and motion data of the robotic arm as input and output a predicted result of the state vector at the next moment. In one embodiment, the input data of the dynamic forward model includes state data, motion data, and module position labels. In another embodiment, the input data of the dynamic forward model may also include module position labels. In one example, the state data may include at least one of barometric pressure sensor and tactile sensor data. In another example, the state data may also include at least one of the pose, axial extension, stiffness characterization value, etc., of the end effector and each segment of the robotic arm. In one example, the motion data may include at least one of cable drive amount and barometric pressure regulation amount. In one example, at least one of the state data and motion data may be in vector form. In one example, the module position labels can be used to characterize the relative order of each segment of the retractable variable stiffness continuum robotic arm in the continuum structure, so that the dynamic forward model can distinguish the individual dynamic characteristics of different segments. The dynamic forward model according to embodiments of this disclosure may be based on a temporal neural network, which may employ a sequence model architecture capable of processing multidimensional temporal data and may optionally employ a network structure with memory units. In one example, the temporal neural network can employ a multi-encoder fusion structure to process heterogeneous inputs such as position, draw line position and force, robotic arm posture, stiffness, air pressure, and tactile sensation, and then input the fused features into the temporal backbone network.

[0034] In step S200, during the current prediction period, based on the task phase of the scalable variable stiffness continuum robot arm and the tactile data from the tactile sensor, the weights corresponding to the multiple terms included in the comprehensive loss function are dynamically adjusted.

[0035] In one example, the multiple terms included in the comprehensive loss function include at least one of an end-effector pose loss term, a telescoping loss term, a stiffness loss term, a velocity loss term, an obstacle avoidance penalty term, and a motion smoothing penalty term. In one embodiment, when the task phase corresponds to an approach phase in which the scalable variable stiffness continuum manipulator approaches a target, the weight of the term associated with the position of the scalable variable stiffness continuum manipulator among the multiple terms included in the comprehensive loss function may be increased. In one embodiment, when the task phase corresponds to an operation phase in which the scalable variable stiffness continuum manipulator performs an operation on a target, the weight of the term associated with the stiffness of the scalable variable stiffness continuum manipulator among the multiple terms included in the comprehensive loss function may be increased. In one embodiment, when a collision is detected via haptic data, the weight of the term associated with the obstacle avoidance penalty among the multiple terms included in the comprehensive loss function may be increased. In one embodiment, the equivalent radius of the virtual obstacle may also be dynamically adjusted to achieve soft obstacle avoidance.

[0036] In step S300, during the current prediction period, the set of state data that minimizes the weighted comprehensive loss function is determined from multiple sets of state data for the next time step. According to embodiments of this disclosure, the target state or target trajectory of the robotic arm can be pre-planned or determined. The set of state data that minimizes the weighted comprehensive loss function can be determined based on the target state and each of the multiple sets of state data for the next time step. In one example, step S300 can be implemented via a learnable policy network, where a dynamic forward model can be used as an environment simulator, taking the current state and target task of the scalable variable stiffness continuum robotic arm as input, and the negative value of the comprehensive loss function as a reward signal, and training the policy network through offline or online reinforcement learning to output corresponding motion data.

[0037] In step S400, during the current control period corresponding to the current prediction period, based on a set of motion data corresponding to a determined set of state data, the extendable variable stiffness continuous body manipulator is driven to perform the motion corresponding to that set of motion data. In one example, the current control period may at least partially overlap with the next prediction period.

[0038] In an optional embodiment, in addition to steps S100 to S400, the data-driven control method for a retractable variable stiffness continuous body manipulator according to the embodiments of the present disclosure may further include: in response to tactile data exceeding a safety threshold, stopping the retractable variable stiffness continuous body manipulator from performing the action corresponding to the set of motion data, and driving the retractable variable stiffness continuous body manipulator to perform a retraction action.

[0039] In an optional embodiment, in addition to steps S100 to S400, the data-driven control method for a scalable variable stiffness continuous body manipulator according to embodiments of the present disclosure may further include: training a dynamic forward model. Hereinafter, reference will be made to... Figure 2 Describe in detail the training of the dynamic feedforward model.

[0040] The data-driven control method for a retractable variable stiffness continuous body robotic arm according to embodiments of this disclosure can be applied to retractable variable stiffness continuous body robotic arms used for home services. For example, a retractable variable stiffness continuous body robotic arm according to embodiments of this disclosure can be deployed on a wheeled mobile platform for kitchen operations. The robotic arm needs to pick up a cup from a table and place it in a sink. During movement, the end effector remains horizontal and has high stiffness to overcome gravity; when approaching the sink, the stiffness decreases and tactile feedback is used to detect edges; if a hand suddenly appears, tactile feedback triggers an emergency stop. Using the control method of this invention, the loss function weights can be automatically switched, which helps to complete picking, placing, and avoidance in complex environments.

[0041] The data-driven control method for a telescopic variable stiffness continuous robotic arm according to embodiments of this disclosure can be applied to telescopic variable stiffness continuous robotic arms used for biological sampling in livestock farming. For example, the telescopic variable stiffness continuous robotic arm according to embodiments of this disclosure can be deployed on a overhead inspection system in a modern livestock farm for non-contact or minimal-contact biological sampling of livestock (such as fecal swabs). During the operation, the livestock are non-cooperative, and their movements are random and unpredictable. Using the control method of this invention, the robotic arm performs sampling tasks in three stages: Approach stage: The robotic arm moves along the rail to the vicinity of the livestock. At this time, the controller increases the position tracking weight, and the end effector quickly approaches the target sampling point while maintaining moderate stiffness to resist minor disturbances. Contact sampling stage: When the tactile sensor detects contact with the livestock's surface or sampling area, the controller automatically switches weights, reducing stiffness and speed weights and increasing tactile obstacle avoidance weights, making the robotic arm compliant and avoiding stress or injury to the livestock. The sampling action is completed through the end effector, and the robotic arm can adjust its posture in real time according to the slight movements of the livestock. Retreat stage: After sampling is completed, the robotic arm increases stiffness and quickly retreats to avoid collisions with the livestock. When a sudden, large movement of the livestock causes an unexpected collision risk, the tactile sensor triggers a safety retraction action (releasing the cable and unloading the air pressure) to ensure that neither the robotic arm nor the livestock is damaged.

[0042] Based on this, the data-driven control method for the scalable variable stiffness continuum manipulator according to the embodiments of this disclosure can control the scalable variable stiffness continuum manipulator with dynamic characteristics that exhibit strong nonlinearity, time-varying, hysteresis and strong correlation with the current state. This enables high-precision, multi-objective (pose, extension, stiffness, speed) collaborative control of the scalable variable stiffness continuum manipulator, adapts to complex task scenarios requiring online stiffness adjustment, speed control and active obstacle avoidance, and improves data efficiency and model generalization.

[0043] Figure 2 This is a flowchart illustrating a method for training a dynamic forward model according to an embodiment of the present disclosure.

[0044] Reference Figure 2 The method for training a dynamic forward model according to embodiments of the present disclosure may include at least step S510.

[0045] In step S510, dynamic response data acquired by applying at least one of a sweep frequency excitation signal, a step excitation signal, and a random combination of excitation signals to a stretchable variable stiffness continuum manipulator can be used as preliminary training data to perform preliminary training on the dynamic forward model. In one example, step S510 can be performed multiple times until the prediction accuracy of the dynamic forward model reaches a preset requirement.

[0046] In an optional embodiment, the method for training a dynamic feedforward model according to embodiments of the present disclosure may further include step S520. In step S520, the dynamic feedforward model may be retrained. In one example, step S520 may be performed multiple times until the prediction accuracy of the dynamic feedforward model reaches a preset requirement. By performing step S520, the prediction accuracy of the trained dynamic feedforward model can be improved. Step S520 may include steps S521 to S524.

[0047] In step S521, multiple value range regions can be divided based on the combination of the value ranges of the state data and the value ranges of the action data. In one example, multiple value range regions can be divided by appropriately discretizing the combination of value ranges.

[0048] In step S522, state training data and action training data corresponding to each of the multiple value range regions can be used to predict the state prediction results corresponding to each of the multiple value range regions based on the pre-trained dynamic forward model.

[0049] In step S523, a confidence level corresponding to each of the plurality of value range regions can be determined based on the state prediction result and the target state. In one example, the confidence level can also be calculated instead of the uncertainty (e.g., model ensemble, Bayesian approximation, etc.).

[0050] In step S524, for the value range corresponding to a confidence level below the confidence threshold, new dynamic response data acquired by applying an excitation signal superimposed with random perturbations to the scalable variable stiffness continuum manipulator can be used as retraining data to retrain the dynamic forward model. In one example, the retraining of the dynamic forward model for the value range corresponding to a confidence level not lower than the confidence threshold can be omitted.

[0051] In an optional embodiment, in addition to steps S100 to S400, the data-driven control method for a scalable variable stiffness continuous body manipulator according to embodiments of the present disclosure may further include: during the current control period, collecting air pressure regulation data and the corresponding response data of the scalable variable stiffness continuous body manipulator as historical air pressure-response data. In this case, in an optional embodiment, the method for training a dynamic forward model according to embodiments of the present disclosure may further include step S530. In step S530, the historical air pressure-response data may be used periodically to perform additional training on the dynamic forward model. In one example, step S530 may be performed after step S510. In another example, step S530 may be performed after step S520. In one example, the dynamic forward model after additional training may be used in the next prediction period. By performing step S530, the prediction accuracy of the trained dynamic forward model can be improved, enabling the dynamic forward model and control strategy to adapt to dynamic disturbances such as load changes and environmental changes.

[0052] Figure 3 This is a block diagram illustrating a data-driven control system for a telescopic variable stiffness continuous body robotic arm according to an embodiment of the present disclosure.

[0053] Reference Figure 3 According to embodiments of the present disclosure, a data-driven control system 100 for a telescopic variable stiffness continuous body robotic arm includes a prediction module 110, a weight adjustment module 120, a determination module 130, and an execution module 140.

[0054] According to an embodiment of this disclosure, the prediction module 110 is configured to: during the current prediction period, using the current state data and multiple sets of motion data of the scalable variable stiffness continuous body manipulator, based on a trained dynamic forward model, predict the next set of state data of the scalable variable stiffness continuous body manipulator corresponding to the multiple sets of motion data respectively.

[0055] According to an embodiment of this disclosure, the weight adjustment module 120 is configured to dynamically adjust multiple weights corresponding to multiple terms included in the comprehensive loss function, based on the task phase of the scalable variable stiffness continuum robot arm and the tactile data of the tactile sensor, during the current prediction period.

[0056] According to an embodiment of this disclosure, the determining module 130 is configured to: during the current prediction period, determine from multiple sets of state data that minimize the weighted comprehensive loss function.

[0057] According to an embodiment of the present disclosure, the execution module 140 is configured to: in the current control period corresponding to the current prediction period, drive the stretchable variable stiffness continuous body robot to perform the action corresponding to the set of action data based on a set of action data corresponding to a determined set of state data.

[0058] Based on this, the data-driven control system 100 of the scalable variable stiffness continuum manipulator according to the embodiments of this disclosure can control the scalable variable stiffness continuum manipulator with dynamic characteristics that exhibit strong nonlinearity, time-varying, hysteresis and strong correlation with the current state. This enables high-precision, multi-target (pose, extension, stiffness, speed) collaborative control of the scalable variable stiffness continuum manipulator, adapts to complex task scenarios that require online stiffness adjustment, speed control and active obstacle avoidance, and improves data efficiency and model generalization.

[0059] According to embodiments of this disclosure, a computer-readable storage medium may also be provided, wherein when instructions stored in the computer-readable storage medium are executed by at least one processor, the at least one processor causes the processor to perform the data-driven control method for a stretchable variable stiffness continuous body robotic arm according to embodiments of this disclosure. Examples of computer-readable storage media include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disc storage, hard disk drive (HDD), solid-state drive (SSD), card storage (such as multimedia cards, secure digital (SD) cards, or ultra-fast digital (XD) cards), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, and any other device configured to store a computer program and any associated data, data files, and data structures in a non-transitory manner and to provide the computer program and any associated data, data files, and data structures to a processor or computer so that the processor or computer can execute the computer program. The computer program in the aforementioned computer-readable storage medium can run in an environment deployed in computer devices such as clients, hosts, agent devices, servers, etc. Furthermore, in one example, the computer program and any associated data, data files, and data structures are distributed across a networked computer system, such that the computer program and any associated data, data files, and data structures are stored, accessed, and executed in a distributed manner through one or more processors or computers.

[0060] According to embodiments of the present disclosure, a computer program product may also be provided, including computer instructions that, when executed by at least one processor, implement the data-driven control method for a stretchable variable stiffness continuous body manipulator according to embodiments of the present disclosure.

[0061] The data-driven control method and system for a scalable variable stiffness continuum manipulator according to embodiments of the present disclosure can control a scalable variable stiffness continuum manipulator with dynamic characteristics exhibiting strong nonlinearity, time-varying, hysteresis, and strong correlation with the current state. This enables high-precision, multi-objective (pose, extension, stiffness, velocity) collaborative control of the scalable variable stiffness continuum manipulator, adapts to complex task scenarios requiring online stiffness adjustment, velocity control, and active obstacle avoidance, and improves data efficiency and model generalization.

[0062] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0063] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A data-driven control method for a telescopic variable stiffness continuous body robotic arm, characterized in that, The data-driven control method includes: During the current prediction period, using the current state data and multiple sets of motion data of the scalable variable stiffness continuous body manipulator, based on the trained dynamic forward model, the next set of state data of the scalable variable stiffness continuous body manipulator corresponding to the multiple sets of motion data is predicted. During the current prediction period, based on the task phase of the scalable variable stiffness continuum robot arm and the tactile data from the tactile sensor, the weights corresponding to the multiple terms included in the comprehensive loss function are dynamically adjusted. During the current prediction period, determine the set of state data that minimizes the weighted comprehensive loss function from the multiple sets of state data for the next time period; During the current control period corresponding to the current prediction period, based on a set of motion data corresponding to a determined set of state data, the telescopic variable stiffness continuous body robotic arm is driven to perform the motion corresponding to the set of motion data.

2. The data-driven control method according to claim 1, characterized in that, The data-driven control method further includes: training a dynamic forward model. The steps for training the dynamic forward model include: The dynamic forward model is initially trained using dynamic response data acquired by applying at least one of a sweep excitation signal, a step excitation signal, and a random combination of excitation signals to a scalable variable stiffness continuum manipulator.

3. The data-driven control method according to claim 2, characterized in that, The steps for training a dynamic forward model also include: Based on the combination of the value range of state data and the value range of action data, multiple value range regions are divided; Using state training data and action training data corresponding to each of the plurality of value range regions, predict the state prediction results corresponding to each of the plurality of value range regions based on the pre-trained dynamic forward model; Based on the state prediction results and the target state, determine the confidence level corresponding to each of the multiple value range regions; For the range of values ​​corresponding to confidence levels below the confidence threshold, new dynamic response data acquired by applying an excitation signal superimposed with random perturbations to a scalable variable stiffness continuum manipulator is used as retraining data to retrain the dynamic forward model.

4. The data drive control method according to claim 2, wherein The data-driven control method further includes: During the current control period, air pressure regulation data and the corresponding response data of the extendable variable stiffness continuous body manipulator are collected as historical air pressure-response data. The steps for training the dynamic forward model also include: The dynamic forward model is periodically trained using historical barometric pressure-response data.

5. The data drive control method of claim 1, wherein During the current prediction period, the steps of dynamically adjusting multiple weights corresponding to various terms included in the comprehensive loss function, based on the task phase of the scalable variable stiffness continuum robot and tactile data from the tactile sensors, include: When the task phase corresponds to the approach phase in which the scalable variable stiffness continuum manipulator approaches the target, increase the weight of the term associated with the position of the scalable variable stiffness continuum manipulator among the multiple terms included in the comprehensive loss function. When the task phase corresponds to the operation phase in which the scalable variable stiffness continuum manipulator performs an operation on the target, increase the weight of the term associated with the stiffness of the scalable variable stiffness continuum manipulator among the multiple terms included in the comprehensive loss function. When a collision is detected via tactile data, the weight of the term associated with obstacle avoidance penalty, which is included among the multiple terms in the comprehensive loss function, is increased.

6. The data drive control method of claim 1, wherein The data-driven control method further includes: In response to the tactile data exceeding a safety threshold, the system stops driving the retractable variable stiffness continuous body robotic arm to perform the action corresponding to the set of motion data, and instead drives the retractable variable stiffness continuous body robotic arm to perform a retraction action. The dynamic forward model is a temporal neural network with memory units that can process multi-dimensional temporal data. Furthermore, the dynamic forward model employs a multi-encoder fusion structure capable of handling heterogeneous inputs such as position, cable position and force, robotic arm posture, stiffness, air pressure, and tactile sensation. The step of determining the set of state data that minimizes the weighted comprehensive loss function from the multiple sets of state data at the next time step is implemented through a learnable policy network. The dynamic forward model is used as an environment simulator, taking the current state and target task of the scalable variable stiffness continuum manipulator as inputs and the negative value of the comprehensive loss function as a reward signal. The policy network is trained through offline or online reinforcement learning to output the corresponding action data.

7. The data-driven control method according to claim 1, characterized in that, The input data for the dynamic forward model includes state data, action data, and module position labels. Status data includes at least one of barometric pressure sensor and tactile sensor data. Motion data includes at least one of cable drive amount and air pressure regulation amount. Module position labels are used to characterize the relative order of the segments of the scalable variable stiffness continuum robot arm within the continuum structure, enabling the dynamic forward model to distinguish the individual dynamic characteristics of different segments. The multiple terms included in the comprehensive loss function include at least one of the following: end-effector pose loss term, stretching loss term, stiffness loss term, velocity loss term, obstacle avoidance penalty term, and motion smoothing penalty term. The current control period overlaps at least partially with the next forecast period.

8. A data-driven control system for a scalable variable-stiffness continuum manipulator, characterized by, The data-driven control system includes: The prediction module is configured to: during the current prediction period, use the current state data and multiple sets of motion data of the scalable variable stiffness continuous body manipulator, based on a trained dynamic forward model, to predict the next set of state data of the scalable variable stiffness continuous body manipulator corresponding to the multiple sets of motion data. The weight adjustment module is configured to dynamically adjust multiple weights corresponding to multiple terms included in the comprehensive loss function based on the task stage of the scalable variable stiffness continuum robot and the tactile data of the tactile sensor during the current prediction period. The determination module is configured to: during the current prediction period, determine the set of state data that minimizes the weighted comprehensive loss function from the multiple sets of state data at the next time step; The execution module is configured to: during the current control period corresponding to the current prediction period, drive the retractable variable stiffness continuous body robot to perform actions corresponding to the set of action data based on a set of action data corresponding to a determined set of state data.

9. A computer-readable storage medium, characterized in that, When the instructions stored in the computer-readable storage medium are executed by at least one processor, the at least one processor is prompted to perform the data-driven control method described above for a scalable variable stiffness continuous body manipulator.

10. A computer program product, characterised in that, The computer program product includes computer instructions that, when executed by at least one processor, implement the aforementioned data-driven control method for a retractable variable stiffness continuous body robotic arm.

Images