Physical Human-Robot Interaction with a Lightweight, Elastic Tendon Driven Robotic Arm

Humans have since long desired to be assisted by robotic systems in productive and home environments. To fulfill this need, efforts are made to increase the cognitive abilities that robots lack to autonomously interpret their environment and human intentions. But equally important, new hardware and actuation designs are required to increase the safety and sensitivity of robots that operate in the vicinity of humans. A main restriction of most current robot arm designs for physical human-robot interaction (pHRI) is the discrepancy of safety and dynamic performance in terms of, for instance, velocity and payload. This thesis therefore deals with the challenges involved in the development of fast robot arms that are safe for the operation in human-centered environments and for applications requiring close pHRI. It presents design guidelines for lightweight robot arms with elastic tendon actuation and, additionally, suitable methods for dynamic modeling and control and safety evaluation. This novel type of robotic arm aims at enabling automation of applications that combine critically high safety requirements for pHRI with high performance and flexibility demands. The BioRob-X4 robot arm is used as a robotic hardware platform for evaluation of the developed models and methods, which are tested in simulation and validated on the robot hardware. In contrast to other robot arm designs, the actuation principle of the BioRob arm is non-modular in order to enable an extreme lightweight and low-inertia design with high safety and acceleration properties. The use of tendons spanning multiple joints, however, introduces kinematic coupling and the use of extension coil springs to maintain tendon tension and to decouple link and rotor inertia introduces undesirable joint oscillations. These effects have to be modeled accurately to investigate the behavior of the actuators and the whole arm dynamics in theory, simulation, and experiment and to allow for the development and design of model-based algorithms. Therefore, detailed mathematical models for the highly compliant and kinematically coupled tendon actuators and the low inertia link structure are developed and validated against experimentally measured data. The actuation models are analyzed with respect to highly dynamic motions inherent to low inertia link designs. Associated effects such as dynamic and static tendon slackening are discussed and from these considerations, guidelines for shaping the actuator characteristic output curves are derived. State space partitioning of the manipulator is proposed for the formulation of the full robot arm dynamics model. By partitioning the model into three state spaces, the dynamics model of the robot arm can be formulated in joint space by reflecting the model states and parameters to the joint space. The presented approach is generally applicable to tendon-driven robotic arms and, furthermore, helpful in reducing the modeling complexity. The design and hardware constraints of the investigated robot arm demand for the development of specific calibration and filter methods for the joint position and velocity states. Thus, a joint position sensor calibration method and a multilevel switching observer are developed that are both in general applicable to robotic arms with high joint elasticity. Based on the inverse dynamics model and the decoupling of tendon actuators spanning multiple joints we derive a position tracking controller by using the developed state space model segmentation. The proposed observer and control methods are evaluated in simulation and on the robot hardware. A new prediction method for maximum collision and clamping forces based on the current dynamic state of the manipulator and its compliant actuators by monitoring also the potential energy stored in the springs is developed and applied successfully. A worst case safety evaluation considering the possibility of software and hardware failures is performed. In this context, the impact behavior of the elastic tendon actuators is compared to robot arms with backdrivable motors that are either stiffly or elastically coupled to the link and either coupled by tendon to the joint or placed directly in the joint. The theoretical and experimental results presented in this thesis demonstrate the feasibility of constructing fast robotic arms with very high safety properties that are suitable for pHRI and operation in close and direct vicinity of humans. The developed detailed multibody dynamics models are applicable to lightweight manipulator arms with stiff kinematic link chains that are driven by highly elastic tendon actuators