Analysis and Control of Variable Stiffness Robots

This dissertation addresses the analysis and control of variable stiffness actuated robots aiming at precise, sensitive, robust, and dynamic interaction with their environment and especially humans. The focus is the adjustment of impedance properties, namely adjusting the robot equilibrium position, the stiffness, and damping. Highest demands are met by the development of new control concepts and the transfer of state-of-the-art techniques from related fields. Physical interaction is a common task of robotic systems. Aiming at providing service and support for humans, robots need to be able to collaborate with humans and in human environments. The ability to perceive the surrounding and react in a sensitive way is a key factor for robots. Another one is the role of mechanical compliance as it allows to attenuate contact forces and both the environment and the robot can be protected. A step towards this goal is the class of torque controlled robots, where active control generates the compliant features using joint torque sensory information. A further advancement of this technology are the recently developed variable stiffness (VS) robots. There, a main feature is the deliberate introduction of a mechanically variable, elastic element in the robot joints. This passive spring provides the desired compliance properties: it increases the mechanism robustness by absorbing shock impacts and its energy storing capabilities allow to achieve highly dynamic motions. Additionally, the spring stiffness can be adjusted and thereby, the robot can be tuned to the task on a mechanical level. A further important property of the elastic elements is its joint torque sensing capability. In this work new control methods are developed and combined with well-established techniques to exploit the full potential of variable stiffness robots. The research has been conducted at the Institute of Robotics and Mechatronics of the German Aerospace Center (DLR), where an integrated variable stiffness robot, the DLR Hand Arm System, has been developed. An introductory part of this thesis is dedicated to the analysis and modeling of variable stiffness actuators (VSA) which is a basis for the later controller designs. An abstract dynamic model is presented describing the variable stiffness actuators and the integration in the multi-joint robot system. The properties of the model and the nonlinear and varying elastic elements are elaborated. To further deepen the actuator understanding, the bidirectional antagonistic variable stiffness (BAVS) setup is evaluated in a case study. Furthermore, the functionality of the BAVS joint and torque and stiffness properties are analysed. Major results are the existence, transition, and stiffness properties of a helping mode, which allows to fully exploit the torque capabilities of the joint. An analytical model of this mode is developed, which facilitates analysis and design of the joint’s stiffness properties. This model is used to synthesize cam disc-based BAVS setups, which are evaluated in experiments. Such, an insight into VS joint design and realization is given and additionally a contribution to actuator development is provided. The major part of this thesis is dedicated to the development of control algorithms to adjust the impedance properties of robots. The impedance control formulation is a main reason for the success of torque controlled robots. It allows to solve many tasks in an intuitive manner by providing flexible adjustment of the properties of the robot such as equilibrium position, stiffness,