Optimal Control of Intrinsically Compliant Robots

Kurzfassung

The vast majority of robots in industrial environments is designed and modeled as stiff multibody systems. Striving for position accuracy and repeatability, any elasticity is regarded as parasitic effects that poses a source of unwanted oscillatory behavior. The compliance is typically realized via impedance control requiring force/torque sensors in each joint, so-called active compliance. Although robots are often praised to outperform humans, this is only true to a very small extent. In general, mammals are capable of outstanding performances with respect to a large variety of different tasks and have been optimized for them over large periods of time through evolutionary processes. Presumably, the feature to store and release energy enables humans to be superior to robots in terms of highly dynamic motions with high peak velocity outputs. Over the recent years, the concept of intrinsically compliant robots has drawn significant attention in the robotics community. The basic idea is to transfer the biological features inherent in the musculoskeletal system to robotics by introducing elastic transmission elements on joint level. These passively compliant systems are not only expected to be more robust against external shock impacts, but also to come closer to human capabilities in terms of robustness and performance. So-called variable stiffness actuators can not only adjust positioning, but also the joint stiffness in order to emulate mammalian muscle (pre-)tension. Although several mechanical systems are being developed right now, only limited work on controlling these novel devices on a fundamental theoretical basis has been considered. This thesis aims to investigate robotic systems with elastic transmission elements between the motor and link under the objective of maximizing the velocity of the final link. In particular, the exploitation of joint elasticity as a temporary energy storage mechanism is to be inverstigated. Since the constraint on the maximum allowable spring deflection is vital to the systems health and any violation would lead to a permanent damage of the system, this constraint plays an important role throughout the thesis. Moreover, the benefits of elastic/variable stiffness robots versus their rigid counterparts are examined. The aforementioned aspects are investigated using mass spring systems to retrieve (as much as possible) analytical solutions and for the nonlinear counterparts, numerical methods are employed in order to unveil similarities or show the limitations of simplified models.