Passivation of a Hierarchical Whole-Body Controller for Humanoid Robots

Humanoid robots for households and service robotic applications are essentially required to be capable of performing multiple tasks in unstructured, unpredictable environments where the robots share their workspace with humans. In this context, guaranteeing a safe and stable interactive behavior during task execution is of paramount importance for these humanoid robots. For the regulation case, by the use of torque control methods that take the active control of the interactive behavior into account such as compliance control, the passivity of the robot with respect to its connected environment is ensured, which is a necessary condition for a stable interaction between the robot and any passive environment. The state-of-the-art multi-objective compliance controller, that is able to implement a strict hierarchy with an arbitrary number of task levels, applies compliance controllers on individual levels for task execution. However, the null space projections used to achieve task hierarchies inevitably destroy the passivity property of the original compliance controller. In this thesis, the source of activity of the classical controller as well as the decoupled closed-loop system are analyzed. Based on the knowledge of the clarified source of activity, an energy-tank passivation method for the classical controller with a two-level task hierarchy is extended to the case of a task hierarchy with an arbitrary number of task levels. Two variations of the energy-tank passivation approaches are presented: the local-energy-tank approach and the global-energy-tank approach. Formal proofs of passivity of the overall system with both approaches are provided. Proofs of additional approach-specific passivity properties are also given. Simulations with a planar manipulator and experiments with the humanoid robot Rollin’ Justin of the DLR validate the theoretical results.