Shape Servoing Control Of Deformable Linear Object Using Manipulator

Shape servoing control of deformable objects is the basis for improving robot manipulation versatility,expanding robot application fields and other needs,and the key to reducing human time costs and manual operation risks,which has fundamental practical significance and theoretical research value.As the complexity of deformable object models,nominal deviations in robot model,and the unknown nature of the task environment often face many challenges,such as safety and reliability requirements,difficulties in accurate parameter identification,and dynamic and unstructured environmental interference.All these factors make designing deformable object shape servoing control systems very difficult.From the perspective of model-free control,this thesis aims to construct a novel shape servoing control system with engineering value for deformable objects to improve the task perception,autonomous decision optimization,and learning control capability of robots.The details are as follows:For the shape detection and feature extraction issue,a self-organizing mapping-based shape detection method is designed to improve the stability of the 3D shape acquisition and is robust to measurement noise.Next,a feature extraction algorithm based on geometric properties is designed to capture the geometric properties of deformable objects using a fitting algorithm and obtain physically meaningful shape features.Then,a deep learningbased feature extraction network is proposed to extract multiple shape configurations and significantly reduce the feature dimensionality.Finally,a feature extraction algorithm based on invariant moments is proposed,which breaks the inherent invariance of contour moments and is robust to environmental disturbance.A model-free shape servoing control method is proposed,which learns the control autonomously directly from the input/output data of the system,avoiding parameter identification and without making any assumptions on the model.The direct shape control of the deformable object is converted to feature control in the hidden space using feature extraction,and the non-linear relationship between shape feature differentiation and robot velocity is simplified to a local linear relationship by introducing the deformation Jacobian matrix,and the rationality of designing the servoing controller on this basis is explained.A shape servoing control method based on model-free adaptive control is proposed for the shape vision servoing control problem in the case of no prior system knowledge.Firstly,a deformation Jacobian matrix online estimation strategy based on state estimation is designed,combined with the Kalman filter algorithm,to achieve the optimal estimation of the deformation Jacobian matrix with strong robustness to measurement noise.After that,a model-free adaptive velocity controller based on an improved criterion function is designed to accelerate the convergence of the system in the presence of disturbance.Then,a control gain adaptive law based on the gradient descent method is designed to regulate the system’s dynamic performance during manipulation,and the asymptotic stability of the system is analyzed using the Lyapunov theory.A shape servoing control method based on sliding mode control is proposed for dynamic manipulation and finite-time convergence problems.First,a shape vision servoing control method based on linear sliding mode control is designed to achieve dynamic manipulation tasks.The stability of the closed-loop system,including identification and control,is analyzed using the Lyapunov theory.After that,a shape servoing control method based on finite-time sliding mode control is designed to adjust the system’s dynamic performance without changing the control gain.The finite-time convergence of the system is analyzed using the Lyapunov theory.A shape servoing control method considering constraints is proposed for solving occlusion,and performance and safety constraints.First,a shape prediction network based on generative adversarial networks is designed to solve the problem of incomplete shape acquisition caused by camera field-of-view occlusion.Then,a numerical algorithm-based online estimation strategy is designed,which reduces the complexity of parameter rectification and owns fast superlinear convergence characteristic.Further,the receding window approach-based online estimation strategy is designed to ensure the deformation Jacobian matrix estimation’s accuracy,smoothness,and low singularity.Then,a model predictive controller based on predicted features is designed to smooth the deformation trajectory,and the system’s asymptotic stability is analyzed using Lyapunov theory.Further,a model predictive controller based on an optimization algorithm is designed,which can explicitly handle the constraint problems and improve the manipulation safety in the constrained environment.