Towards Lifelong Learning of Optimal Control for Kinematically Complex Robots
Alexander Dettmann, Malte Langosz, Kai von Szadkowski, Sebastian Bartsch
In ICRA14 Workshop on Modelling, Estimation, Perception and Control of All Terrain Mobile Robots, (ICRA-2014), 31.5.-07.6.2014, Hong Kong, IEEE, Jun/2014.
Robots intended to perform mobile manipulation in complex environments are commonly equipped with an extensive set of sensors and motors, creating a wide range of perception and interaction capabilities. However, to exploit all theoretically possible abilities of such systems, a control strategy is required that allows to determine and apply the best solution for a given task within an appropriate time frame. In this paper, a lifelong self-improving control scheme for kinematically complex robots is presented, which uses simulation-based behavior generation and optimization procedures to create a library of well-performing solutions for varying tasks and conditions, and combines it with case-based selection, evaluation, and online adaptation methods.