<p>Industrial robots have been used for a long time in the industry. Despite this thedevelopment of advanced force control system using industrial robots is relativelylimited. Using force controlled robot systems expands the possibility of what canbe done with industrial robots.Previously a force feedback system for a standard industrial robot from ABBhas been developed. The system is developed towards the aircraft industry, where amounted drill machine on the robot has to fulfill the requirements in robot drillingin aircraft structures. This thesis presents experimental results and improvementsof this industrial robot system. Mechanical modifications and tests of a new endeffector are analyzed.</p>

The motor control of anthropomorphic robotic systems is a challenging computational task mainly because of the high levels of redundancies such systems exhibit. Optimality principles provide a general strategy to resolve such redundancies in a task driven fashion. In particular closed loop optimisation, i.e., optimal feedback control (OFC), has served as a successful motor control model as it unifies important concepts such as costs, noise, sensory feedback and internal models into a coherent mathematical framework. Realising OFC on realistic anthropomorphic systems however is non-trivial: Firstly, such systems have typically large dimensionality and nonlinear dynamics, in which case the optimisation problem becomes computationally intractable. Approximative methods, like the iterative linear quadratic gaussian (ILQG), have been proposed to avoid this, however the transfer of solutions from idealised simulations to real hardware systems has proved to be challenging. Secondly, OFC relies on an accurate description of the system dynamics, which for many realistic control systems may be unknown, difficult to estimate, or subject to frequent systematic changes. Thirdly, many (especially biologically inspired) systems suffer from significant state or control dependent sources of noise, which are difficult to model in a generally valid fashion. This thesis addresses these issues with the aim to realise efficient OFC for anthropomorphic manipulators. First we investigate the implementation of OFC laws on anthropomorphic hardware. Using ILQG we optimally control a high-dimensional anthropomorphic manipulator without having to specify an explicit inverse kinematics, inverse dynamics or feedback control law. We achieve this by introducing a novel cost function that accounts for the physical constraints of the robot and a dynamics formulation that resolves discontinuities in the dynamics. The experimental hardware results reveal the benefits of OFC over traditional (open loop) optimal controllers in terms of energy efficiency and compliance, properties that are crucial for the control of modern anthropomorphic manipulators. We then propose a new framework of OFC with learned dynamics (OFC-LD) that, unlike classic approaches, does not rely on analytic dynamics functions but rather updates the internal dynamics model continuously from sensorimotor plant feedback. We demonstrate how this approach can compensate for unknown dynamics and for complex dynamic perturbations in an online fashion. A specific advantage of a learned dynamics model is that it contains the stochastic information (i.e., noise) from the plant data, which corresponds to the uncertainty in the system. Consequently one can exploit this information within OFC-LD in order to produce control laws that minimise the uncertainty in the system. In the domain of antagonistically actuated systems this approach leads to improved motor performance, which is achieved by co-contracting antagonistic actuators in order to reduce the negative effects of the noise. Most importantly the shape and source of the noise is unknown a priory and is solely learned from plant data. The model is successfully tested on an antagonistic series elastic actuator (SEA) that we have built for this purpose. The proposed OFC-LD model is not only applicable to robotic systems but also proves to be very useful in the modelling of biological motor control phenomena and we show how our model can be used to predict a wide range of human impedance control patterns during both, stationary and adaptation tasks.

abstract: As robotic systems are used in increasingly diverse applications, the interaction of humans and robots has become an important area of research. In many of the applications of physical human robot interaction (pHRI), the robot and the human can be seen as cooperating to complete a task with some object of interest. Often these applications are in unstructured environments where many paths can accomplish the goal. This creates a need for the ability to communicate a preferred direction of motion between both participants in order to move in coordinated way. This communication method should be bidirectional to be able to fully utilize both the robot and human capabilities. Moreover, often in cooperative tasks between two humans, one human will operate as the leader of the task and the other as the follower. These roles may switch during the task as needed. The need for communication extends into this area of leader-follower switching. Furthermore, not only is there a need to communicate the desire to switch roles but also to control this switching process. Impedance control has been used as a way of dealing with some of the complexities of pHRI. For this investigation, it was examined if impedance control can be utilized as a way of communicating a preferred direction between humans and robots. The first set of experiments tested to see if a human could detect a preferred direction of a robot by grasping and moving an object coupled to the robot. The second set tested the reverse case if the robot could detect the preferred direction of the human. The ability to detect the preferred direction was shown to be up to 99% effective. Using these results, a control method to allow a human and robot to switch leader and follower roles during a cooperative task was implemented and tested. This method proved successful 84% of the time. This control method was refined using adaptive control resulting in lower interaction forces and a success rate of 95%. / Dissertation/Thesis / M.S. Mechanical Engineering 2014

Opening a door is still a hard problem in robotics. Many robotic manipulators use open-loop position control to open doors, which reduces reusability and reliability in the face of slight differences or sensor errors. Many others use force feedback or impedance control but skip past the problem of grabbing the handle, which could lead to failures due to sensor errors. This research assumes that perception is faulty, and uses joint-level force feedback to probe the location of the door and its handle before attempting to open it. The resulting control strategy is at least 33% faster than the open-loop control system it replaces, and had an 83% success rate during testing in place of the previous method's 60% success rate. / Master of Science

Esta tese trata da estimativa de torque e rigidez do paciente dirigida por EMG e sua utilização para adaptar a rigidez do robô durante a reabilitação assistida por robôs. Os sinais eletromiográficos (EMG), obtidos de músculos que atuam durante os movimentos de flexão e extensão de um usuário utilizando uma órtese de joelho ativa, são processados para obter as ativações dos músculos. Inicialmente, um modelo musculoesquelético simplificado e otimizado é usado para calcular as estimativas de torque e rigidez da junta do paciente. A otimização do modelo é realizada comparando o torque estimado com o torque gerado pela ferramenta de dinâmica inversa do software OpenSim, considerando um modelo musculoesquelético escalonado. Como uma solução complementar, é proposta uma rede neural perceptron multicamada (NN) para mapear os sinais EMG para o torque do paciente. Também é apresentado um Ambiente de Estimativa de Torque Gerado por EMG criado para analisar os dados obtidos a partir da aplicação das abordagens propostas considerando a aplicação de um protocolo criado para a análise de interação usuário-exoesqueleto. Um banco de dados de indivíduos saudáveis também foi disponibilizado neste trabalho. Além disso, uma estratégia de controle de impedância adaptativa é proposta para ajustar a rigidez do robô com base na estimativa de rigidez do paciente por EMG. A estratégia inclui uma solução ideal para a interação paciente-robô. Finalmente, são apresentados os resultados obtidos aplicando o controle de impedância adaptativo proposto durante os movimentos de flexão e extensão do usuário que usa a órtese ativa. / This thesis deals with EMG-driven patient torque and stiffness estimation and its use to adapt the robot stiffness during robot-aided rehabilitation. Electromyographic (EMG) signals, taken from selected muscles acting during flexion and extension movements of an user wearing an active knee orthosis, are processed to get the muscles activations. First, a simplified and optimized musculoskeletal model is used to compute the estimate of patient joint torque and stiffness. The model optimization is performed by comparing the estimate torque with the torque generated by the inverse dynamics tool of the OpenSim software, considering a scaled musculoskeletal model. As a complementary solution, a multilayer perceptron neural network (NN) is proposed to map the EMG signals to the patient torque. It is also presented an EMG-driven Torque Estimation Environment created to analyze the data obtained from the application of the proposed approaches considering a protocol created for user-exoskeleton interaction analysis. A database with data from 5 healthy subjects is also made available in this work. Additionally, an adaptive impedance control strategy is proposed to adjust the robot stiffness based on the EMG-driven patient stiffness estimation. The strategy includes an optimal solution for the patient-robot interaction. Finally, the results obtained by applying the proposed adaptive impedance control during flexion and extension movements of the user wearing the active orthosis are presented.

Esta dissertação apresenta o projeto e a construção de atuadores elásticos em série para o acionamento das juntas de um exoesqueleto para membros inferiores, baseado em uma órtese comercial. Inicialmente, considerou-se como dispositivo de testes a parte do exoesqueleto referente à junta do tornozelo, ou seja, a construção de uma órtese tornozelo-pé ativa. Atuadores elásticos em série são considerados neste trabalho, pois tais dispositivos apresentam características ideais para a sua utilização em órteses ativas: controle de força, controle de impedância (possibilidade de impedância baixa), absorção de impactos, baixo atrito e largura de banda que se aproxima da movimentação muscular. Um primeiro protótipo do atuador elástico em série foi construído e resultados experimentais de controle de força, impedância e posição foram obtidos com sucesso, através de uma interface de acionamento e controle entre o atuador, os sensores (encoders e sensores de força) e o computador. Também foi construída uma órtese tornozelo-pé ativa acionada pelo atuador elástico em série construído, sendo apresentados os primeiros resultados experimentais obtidos com este dispositivo. / This dissertation deals with the design and construction of series elastic actuators for driving the joints of an exoskeleton for lower limbs, based on a commercial orthosis. Initially, it was considered the construction of the exoskeleton\'s ankle joint, that is, the construction of an active ankle-foot orthosis. Series elastic actuators are considered in this work since these devices have ideal characteristics for use in active orthoses: force control, impedance control (possibility of low impedance), impact absorption, low friction and bandwidth that approximates the muscle movement. A first prototype of the series elastic actuator was constructed and experimental results of force, impedance, and position control were successfully obtained trough of a control interface between the actuators, the sensors (encoders and force sensors) and the computer. Also, an active ankle-foot orthosis, driven by the series elastic actuator, was constructed and the first experimental results achieved with this device are presented.

The da Vinci Surgical System is one of the most established robot-assisted surgery device commended for its dexterity and ergonomics in minimally invasive surgery. Conversely, it inherits disadvantages which are lack of autonomy and haptic feedback. In order to address these issues, this work proposes an industry-inspired solution to the field of force control in medical robotics. This approach contributes to shared autonomy by developing a controller for cooperative object manipulation with force tracking utilizing available manipulators and force feedback. To achieve simultaneous position and force tracking of the object, master and slave manipulators were assigned then controlled with Cartesian position control and impedance control respectively. Because impedance control requires a model-based feedforward compensation, we identified the lumped base parameters of mass, inertias, and frictions of a three degree-of-freedom double four-bar linkage mechanism with least squares and weighted least squares regression methods. Additionally, semidefinite programming was used to constrain the parameters to a feasible physical solution in standard parameter space. Robust stick-slip static friction compensation was applied where linear Viscous and Coulomb friction was inadequate in modeling the prismatic third joint. The Robot Operating System based controller was tested in RViz to check the cooperative kinematics of up to three manipulators. Additionally, simulation with the dynamic engine Gazebo verified the cooperative controller applying a constant tension force on a massless spring-damper virtual object. With adequate model feedback linearization, the cooperative impedance controller tested on the da Vinci Research Kit yielded stable tension force tracking while simultaneously moving in Cartesian space. The maximum force tracking error was +/- 0.5 N for both a compliant and stiff manipulated object.

Este trabalho apresenta o desenvolvimento de uma estratégia de assistência adaptativa mediante a implementação de um controle de impedância variável para um robô de reabilitação do tornozelo. A estratégia é formulada de tal forma que o dispositivo robótico assiste ao paciente somente quando e quanto for necessário, seguindo o paradigma Assist-As-Needed. Inicialmente, a contribuição dinâmica do paciente durante a realização do movimento é estimada com base nas informações cinemáticas e de torque fornecidas pelo robô. Em seguida, são propostos dois métodos para se obter o parâmetro de rigidez do controlador de impedância, o primeiro deles determina um valor de erro admissível e calcula a rigidez do robô para complementar a atuação do paciente, e o segundo calcula a rigidez mediante a minimização de um funcional que quantifica o processo de reabilitação e a interação entre robô e paciente. Além disso, a quantidade de assistência dada pelo robô também é adaptada conforme o desempenho do paciente ao longo da sessão. A estratégia foi implementada no robô Anklebot e avaliada em três pacientes pós-AVC para movimentos de flexão dorsal/plantar e de inversão/eversão. Os resultados obtidos indicam que o método utilizado para a estimativa da rigidez é válido para determinar a quantidade de assistência. Finalmente, os resultados confirmam que o aumento do desempenho do paciente gera uma diminuição da assistência robótica, e vice-versa. / This work presents the design of an adaptive robotic assistance strategy through a variable impedance control of an ankle rehabilitation robot. This strategy is formulated so that the robotic device assists the patient only as much as needed, following the Assist-As-Needed paradigm. First, the dynamic contribution of the patient during the motion is estimated based on the torque and kinematic information provided by the robot. Then, two methods are proposed to calculate the stiffness parameter of the impedance controller, the first one determines an admisible value of error and computes the robot stiffness to complement the estimated patient stiffness. The second one computes the robot stiffness by minimizing a functional that quantifies both the rehabilitation process and the interaction between robot and patient. In addition, the amount of the robotic assistance is adapted according to the patient\'s performance. The proposed methods were implemented at the Anklebot and evaluated by three post-stroke patients for dorsi/plantarflexion and inversion/eversion movements. Results indicate that the stiffness estimation is a valid method to determine the amount of the assistance. Finally, the results confirm that increasing the performance of the patient generates a decrease in the robotic assistance, and vice versa.

Esta tese trata da estimativa de torque e rigidez do paciente dirigida por EMG e sua utilização para adaptar a rigidez do robô durante a reabilitação assistida por robôs. Os sinais eletromiográficos (EMG), obtidos de músculos que atuam durante os movimentos de flexão e extensão de um usuário utilizando uma órtese de joelho ativa, são processados para obter as ativações dos músculos. Inicialmente, um modelo musculoesquelético simplificado e otimizado é usado para calcular as estimativas de torque e rigidez da junta do paciente. A otimização do modelo é realizada comparando o torque estimado com o torque gerado pela ferramenta de dinâmica inversa do software OpenSim, considerando um modelo musculoesquelético escalonado. Como uma solução complementar, é proposta uma rede neural perceptron multicamada (NN) para mapear os sinais EMG para o torque do paciente. Também é apresentado um Ambiente de Estimativa de Torque Gerado por EMG criado para analisar os dados obtidos a partir da aplicação das abordagens propostas considerando a aplicação de um protocolo criado para a análise de interação usuário-exoesqueleto. Um banco de dados de indivíduos saudáveis também foi disponibilizado neste trabalho. Além disso, uma estratégia de controle de impedância adaptativa é proposta para ajustar a rigidez do robô com base na estimativa de rigidez do paciente por EMG. A estratégia inclui uma solução ideal para a interação paciente-robô. Finalmente, são apresentados os resultados obtidos aplicando o controle de impedância adaptativo proposto durante os movimentos de flexão e extensão do usuário que usa a órtese ativa. / This thesis deals with EMG-driven patient torque and stiffness estimation and its use to adapt the robot stiffness during robot-aided rehabilitation. Electromyographic (EMG) signals, taken from selected muscles acting during flexion and extension movements of an user wearing an active knee orthosis, are processed to get the muscles activations. First, a simplified and optimized musculoskeletal model is used to compute the estimate of patient joint torque and stiffness. The model optimization is performed by comparing the estimate torque with the torque generated by the inverse dynamics tool of the OpenSim software, considering a scaled musculoskeletal model. As a complementary solution, a multilayer perceptron neural network (NN) is proposed to map the EMG signals to the patient torque. It is also presented an EMG-driven Torque Estimation Environment created to analyze the data obtained from the application of the proposed approaches considering a protocol created for user-exoskeleton interaction analysis. A database with data from 5 healthy subjects is also made available in this work. Additionally, an adaptive impedance control strategy is proposed to adjust the robot stiffness based on the EMG-driven patient stiffness estimation. The strategy includes an optimal solution for the patient-robot interaction. Finally, the results obtained by applying the proposed adaptive impedance control during flexion and extension movements of the user wearing the active orthosis are presented.

This dissertation demonstrates a new method for inferring a representation of the motor command, generated by the central nervous system for interactive point-to-point movements. This new tool, the input inference neural network or IINN, allows estimation of the complexity of the motor command. The IINN was applied to experimental data gathered from 7 volunteer subjects who performed point-to-point tasks while interacting with a specially constructed haptic robot. The motor plan inference demonstrates that, for the point-to-point movement tasks executed during experiments, the motor command can be projected onto a low-dimensional manifold. This dimension is estimated to be 4 or 5 and far less than the degrees of freedom available in the arm. It is hypothesized that subjects simplify the problem of adapting to changing environments by projecting the motor control problem onto a motor manifold of low dimension. Reducing the dimension of the movement optimization problem through the development of a motor manifold can explain rapid adaptation to new motor tasks.