Machine Learning and Dynamic Whole Body Control for Underwater Manipulation

Abstract :

Autonomous underwater manipulation is nowadays still an open research challenge. This paper describes the approaches to tackle some of the open challenges. On the one side, the use of machine learning techniques for the online identification and adaption of vehicle dynamics (dealing with drift compensation, mass changes, etc.) as well as the use of high-level context-based configuration of controllers to adapt to changes in system morphology, hardware, and/or tasks. On the other side, a robust control of underwater manipulators based on an extension of whole-body control techniques is envisaged which takes into account the heterogeneous actuation (thrusters on the base, actuators on the arm joints) as well as the uncertain underwater vehicle dynamics. The result is a highly-reconfigurable system that can automatically adapt its behavior to cope with changes in the environment, in its own morphology and/or in the task goals. The outcomes are planned to be validated in two different scenarios: a floating-base dynamics testbed originating from space applications and aerial robots at DLR and an underwater pool at DFKI.

Files:

preprint-chapter.pdf

Links:

https://doi.org/10.1007/978-3-030-30683-0_9