Momentum trajectory generation and control for multi-contact interaction

Legged robots are expected to locomote autonomously in an uncertain and potentially dynamically changing environment. Active interaction with contacts becomes inevitable to move and apply forces in a goal directed way and withstand unpredicted changes in the environment. Therefore, we need to design algorithms that exploit interaction forces and generate desired motions of the robot leading to robust and compliant interaction with the environment. As complexity of tasks increases, generating motion trajectories quickly in combination with feedback control laws that can be executed on the full robot will become increasingly more important and open the way for agile robots. Dynamic motions that require contact interaction with several endeffectors at the same time need to be planned over a time horizon in order to result in trajectories that are realizable on the robot. However, planning with the full robot model can result in excessive computation time, even though only a subset of the dynamics might be required to model the robot behavior well. Execution of planned trajectories, on the other hand, has been shown to work well when inverse dynamics is used under consideration of the full robot dynamics in fast control loops [1]. As a consequence, in this work we address the problem of dynamic motion generation and control by separating it into two different time-scales and levels of modeling granularities: We use trajectory optimization on reduced robot dynamics and generate motion plans over a horizon in a few minutes of optimization. The resulting trajectories are then tracked within a Hierarchical Inverse Dynamics whole-body controller together with other tasks and constraints in a real-time control loop on the kHz level.