Reactive collision avoidance for highly elastic robot systems: an artificial potential field approach

Kurzfassung

Over the past few years, robot design has evolved from rigid towards mechanically com- pliant robots [GASB + 11, DMA + 11]. Higher robustness against impacts is only one amongst many other benefits physical compliance provides [PW95, EHW + 10, WH08]. Nevertheless, collision avoidance plays an important role for a safe interaction between the elastic robot and its environment. Due to their intuitively understandable function- ality, artificial potential fields artificial potential fields (APFs) are commonly used as a collision avoidance method for rigid robots [SASO + 07, Kha85]. This thesis aims at examining the transferability of this method from rigid to highly elastic robots. For this purpose an artificial potential field is designed in joint space for DLR David, a compliantly actuated robot with nonlinear spring characteristics, devel- oped by Deutsches Zentrum für Luft- und Raumfahrt (DLR). It results in a repulsive torque and is implemented by means of Elastic Structure Preserving Impedance (ESfi) control [KLOAS18b]. A multi-body simulation shows that artificial potential fields are appropriate to avoid predefined joint angle limits. This can be considered as collision avoidance in joint space. The challenge is to design an APF such that certain demands are fulfilled. Compared to a rigid robot, a compliant robot and in this case the control method used require additional characteristics of the APF, for instance the continuity of the corresponding torque up to its second derivative. The description of the desired torque meets the re- quirements partly by a function defined piecewisely. In order to prevent oscillations, a damping torque is designed additionally. The shape of the APF including damping can be adjusted by three free parameters of these basic functions. They are tuned by dint of simulation ensuring that the remaining requirements are fulfilled. For that purpose, three situations are considered. In the first and second situation the artificial potential field including damping is applied to only one joint. For the first situation, an auto- mated parameter variation is performed. In the third situation, the artificial potential field including damping is applied to two joints. It is demonstrated that for each situation exists at least one parameter value set (PVS) that ensures that all requirements are fulfilled. Additionally, the proof of concept is supported by the fact that the majority of PVSs, tested in the automated parameter variation, ensures that all requirements are met. Furthermore, the influence of each parameter on the required maximum motor torque and other quantities is evaluated for the first situation. Another important insight is that a parameter optimization offers great potential.