Benchmarking and Validation for Nonlinear Mode Analysis

Kurzfassung

One research focus at the Robotics and Mechatronics Center of the German Aerospace Center in Oberpfaffenhofen is the improvement of robotic locomotion capabilities. Therefore, natural role models are analyzed. As an inspiration, mammalian locomotion is analyzed, which is unmatched in its energy efficiency and adaptility due to the opitmal integration of control strategies and intrinsic body dynamics. In an attempt to recreate this integration in robotic hardware, the compliant quadruped Bert has been developed and built as a test platform. As a contribution towards this goal, this thesis investigates the excitation of nonlinear intrinsic motions of one individual leg of the robot Bert. Springs in the joints of the leg enable dynamic motions. As first approach to model the individual leg in Matlab for the controller design process, an elastic double pendulum is set up. This is followed by the consideration of different modelling methods such as the Pfaffian constraints method and the active joints method. These methods lead to a leg model. For the analysis of the intrinsic dynamics, the concept of nonlinear normal modes is used which was recently explored and published by the German Aerospace Center. After the nonlinear normal modes for the leg are computed, the influence of parameter variation on the shapes of manifold and generators is evaluated. Subsequently, a nonlinear normal mode is found that is a prominate candidate to excite an energy efficient hopping motion in the oneleg of Bert. To initialize and stabilize the modal oscillation, a position mode controller is evolved from an existing torque mode controller. After the control framework is set up, tests are performed in the 3D physics simulation software Gazebo with a digital twin of the Bert leg. The experiments showed that some modifications will be needed for the implementation of the developed position controller to work on the real hardware of the Bert leg. Nevertheless, the simulation experiments also showed that the controller is capable to stabilize modal oscillations with little control effort in an idealized system proving the developed control strategy to be highly energy efficient. As a result, this thesis takes another step towards the use of intrinsic system dynamics in the field of robotics based on natural role models.