No description available.

No description available.

No description available.

No description available.

No description available.

This thesis presents a novel bilateral teleoperation algorithm for n degree of freedom nonlinear manipulators connected through time delays. Teleoperation has many practical uses, as there are many benefits that come from being able to operate machines from a distance. For instance, the ability to send a remote controlled robotic vehicle into a hazardous environment can be a great asset in many industrial applications. As well, the field of remote medicine can benefit from these technologies. A highly skilled surgeon could perform surgery on a patient who is located in another city, or even country. Earth to space operations and deep sea exploration are other areas where teleoperation is quite useful. Central to the approach presented in this work is the use of second order sliding mode unknown input observers for estimating the external forces acting on the manipulators. The use of these observers removes the need for both velocity and force sensors, leading to a lower cost hardware setup that provides all of the advantages of a position-force teleoperation algorithm. Stability results for this new algorithm are presented for several cases. Stability of each of the master and slave sides of the teleoperation system is demonstrated, showing that the master and slave are both stabilized by their respective controllers when the unknown input observers are used for state and force estimation. Additionally, closed loop stability results for the teleoperation system connected to a variety of slave side environments are presented. Delay-independent stability results for a linear spring-damper environment as well as a general finite-gain stable nonlinear environment are given. Delay-dependent stability results for the case where the slave environment is a liner spring-damper and the delays are commensurate are also presented. As well, stability results for the closed loop under the assumption that the human operator is modeled as a finite-gain stable nonlinear environment are given. Following the theoretical presentation, numerical simulations illustrating the algorithm are presented, and experimental results verifying the practical application of the approach are given.

This thesis presents a novel bilateral teleoperation algorithm for n degree of freedom nonlinear manipulators connected through time delays. Teleoperation has many practical uses, as there are many benefits that come from being able to operate machines from a distance. For instance, the ability to send a remote controlled robotic vehicle into a hazardous environment can be a great asset in many industrial applications. As well, the field of remote medicine can benefit from these technologies. A highly skilled surgeon could perform surgery on a patient who is located in another city, or even country. Earth to space operations and deep sea exploration are other areas where teleoperation is quite useful. Central to the approach presented in this work is the use of second order sliding mode unknown input observers for estimating the external forces acting on the manipulators. The use of these observers removes the need for both velocity and force sensors, leading to a lower cost hardware setup that provides all of the advantages of a position-force teleoperation algorithm. Stability results for this new algorithm are presented for several cases. Stability of each of the master and slave sides of the teleoperation system is demonstrated, showing that the master and slave are both stabilized by their respective controllers when the unknown input observers are used for state and force estimation. Additionally, closed loop stability results for the teleoperation system connected to a variety of slave side environments are presented. Delay-independent stability results for a linear spring-damper environment as well as a general finite-gain stable nonlinear environment are given. Delay-dependent stability results for the case where the slave environment is a liner spring-damper and the delays are commensurate are also presented. As well, stability results for the closed loop under the assumption that the human operator is modeled as a finite-gain stable nonlinear environment are given. Following the theoretical presentation, numerical simulations illustrating the algorithm are presented, and experimental results verifying the practical application of the approach are given.

Model-based control strategies for robot manipulators can present numerous performance advantages when an accurate model of the system dynamics is available. In practice, obtaining such a model is a challenging task which involves modeling such physical processes as friction, which may not be well understood and difficult to model. Furthermore, uncertainties in the physical parameters of a system may be introduced from significant discrepancies between the manufacturer data and the actual system. Traditionally, adaptive and robust control strategies have been developed to deal with parametric uncertainty in the dynamic model, but often require knowledge of the structure of the dynamics. Recent approaches to model-based manipulator control involve data-driven learning of the inverse dynamics relationship, eliminating the need for any a-priori knowledge of the system model. Locally Weighted Projection Regression (LWPR) has been proposed for learning the inverse dynamics function of a manipulator. Due to its use of simple local, linear models, LWPR is suitable for online and incremental learning. Although global regression techniques such as Gaussian Process Regression (GPR) have been shown to outperform LWPR in terms of accuracy, due to its heavy computational requirements, GPR has been applied mainly to offline learning of inverse dynamics. More recent efforts in making GPR computationally tractable for real-time control have resulted in several approximations which operate on a select subset, or sparse representation of the entire training data set. Despite the significant advancements that have been made in the area of learning control, there has not been much work in recent years to evaluate these newer regression techniques against traditional model-based control strategies such as adaptive control. Hence, the first portion of this thesis provides a comparison between a fixed model-based control strategy, an adaptive controller and the LWPR-based learning controller. Simulations are carried out in order to evaluate the position and orientation tracking performance of each controller under varied end effector loading, velocities and inaccuracies in the known dynamic parameters. Both the adaptive controller and LWPR controller are shown to have comparable performance in the presence of parametric uncertainty. However, it is shown that the learning controller is unable to generalize well outside of the regions in which it has been trained. Hence, achieving good performance requires significant amounts of training in the anticipated region of operation. In addition to poor generalization performance, most learning controllers commence learning entirely from `scratch,' making no use of any a-priori knowledge which may be available from the well-known rigid body dynamics (RBD) formulation. The second portion of this thesis develops two techniques for online, incremental learning algorithms which incorporate prior knowledge to improve generalization performance. First, prior knowledge is incorporated into the LWPR framework by initializing the local linear models with a first order approximation of the prior information. Second, prior knowledge is incorporated into the mean function of Sparse Online Gaussian Processes (SOGP) and Sparse Pseudo-input Gaussian Processes (SPGP), and a modified version of the algorithm is proposed to allow for online, incremental updates. It is shown that the proposed approaches allow the system to operate well even without any initial training data, and further performance improvement can be achieved with additional online training. Furthermore, it is also shown that even partial knowledge of the system dynamics, for example, only the gravity loading vector, can be used effectively to initialize the learning.

The subject of this thesis is the design and practical application of a model-based controller with variable structure control (VSC). Robot manipulators are highly non-linear systems, however they form a specific class in the non-linear group. Exact mathematical descriptions of the robot dynamics can be achieved and further, robot manipulators have specific useful properties that can be used for the design of advanced controllers. The inclusion of the inverse dynamic description of the robot manipulator as a feedforward term of the controller (model-based controller) is used to transform two non-linear systems i.e. the controller and the robot, into one linear system. The limitation of this technique arises from the accuracy of the inverse dynamic model. The linearisation only takes place if the model is known exactly. To deal with the uncertainties that arise in the model, a control methodology based on variable structure control is proposed. The design of the controller is based on a Lyapunov approach and engineering considerations of the robot. A candidate Lyapunov function of a pseudo-energy form is selected to start the controller design. The general form of the controller is selected to satisfy the negative definiteness of the Lyapunov function. The initial uncertainties between the actual robot dynamics and the model used in the controller are dealt with using a classical VSC regulator. The deficiencies of this approach are evident however because of the chattering phenomenum. The model uncertainties are examined from an engineering point of view and adjustable bounds are then devised for the VSC regulator, and simulations confirm a reduction in the chattering. Implementation on the SPRINTA robot reveals further limitations in the proposed methodology and the bound adjustment is enhanced to take into account the position of the robot and the tracking errors. Two controllers based on the same principle are then obtained and their performances are compared to a PID controller, for three types of trajectory. Tests reveal the superiority of the devised control methodology over the classic PID controller. The devised controller demonstrates that the inclusion of the robot dynamics and properties in the controller design with adequate engineering considerations lead to improved robot responses.

Model-based control strategies for robot manipulators can present numerous performance advantages when an accurate model of the system dynamics is available. In practice, obtaining such a model is a challenging task which involves modeling such physical processes as friction, which may not be well understood and difficult to model. Furthermore, uncertainties in the physical parameters of a system may be introduced from significant discrepancies between the manufacturer data and the actual system. Traditionally, adaptive and robust control strategies have been developed to deal with parametric uncertainty in the dynamic model, but often require knowledge of the structure of the dynamics. Recent approaches to model-based manipulator control involve data-driven learning of the inverse dynamics relationship, eliminating the need for any a-priori knowledge of the system model. Locally Weighted Projection Regression (LWPR) has been proposed for learning the inverse dynamics function of a manipulator. Due to its use of simple local, linear models, LWPR is suitable for online and incremental learning. Although global regression techniques such as Gaussian Process Regression (GPR) have been shown to outperform LWPR in terms of accuracy, due to its heavy computational requirements, GPR has been applied mainly to offline learning of inverse dynamics. More recent efforts in making GPR computationally tractable for real-time control have resulted in several approximations which operate on a select subset, or sparse representation of the entire training data set. Despite the significant advancements that have been made in the area of learning control, there has not been much work in recent years to evaluate these newer regression techniques against traditional model-based control strategies such as adaptive control. Hence, the first portion of this thesis provides a comparison between a fixed model-based control strategy, an adaptive controller and the LWPR-based learning controller. Simulations are carried out in order to evaluate the position and orientation tracking performance of each controller under varied end effector loading, velocities and inaccuracies in the known dynamic parameters. Both the adaptive controller and LWPR controller are shown to have comparable performance in the presence of parametric uncertainty. However, it is shown that the learning controller is unable to generalize well outside of the regions in which it has been trained. Hence, achieving good performance requires significant amounts of training in the anticipated region of operation. In addition to poor generalization performance, most learning controllers commence learning entirely from `scratch,' making no use of any a-priori knowledge which may be available from the well-known rigid body dynamics (RBD) formulation. The second portion of this thesis develops two techniques for online, incremental learning algorithms which incorporate prior knowledge to improve generalization performance. First, prior knowledge is incorporated into the LWPR framework by initializing the local linear models with a first order approximation of the prior information. Second, prior knowledge is incorporated into the mean function of Sparse Online Gaussian Processes (SOGP) and Sparse Pseudo-input Gaussian Processes (SPGP), and a modified version of the algorithm is proposed to allow for online, incremental updates. It is shown that the proposed approaches allow the system to operate well even without any initial training data, and further performance improvement can be achieved with additional online training. Furthermore, it is also shown that even partial knowledge of the system dynamics, for example, only the gravity loading vector, can be used effectively to initialize the learning.