Learning to Control Space Robots with Flexible Appendages Using Model-Based Policy Search

Qiyuan Zhang,Deshan Meng,Xuectian Wang,Bin Liang,Weining Lu
DOI: https://doi.org/10.1109/robio.2017.8324612
2017-01-01
Abstract:Space robot is one of the promising approach to perform space tasks. Modeling and controlling of space robots is complicated caused by high dynamics coupling between the manipulator and the target spacecraft. Beyond that, vibration of flexible appendages mounted on the base of space robot can be excited, which is not expected for the precise control. After capturing the target satellite, the manipulator will be controlled to move from the capturing configuration to the berthing configuration, which is conveniently servicing the target. In this paper, we propose to learn the control policy of space robot directly from observed trajectories. A Gaussian regression model fitted to observed trajectories is used to approximate dynamics of space robotic system, based on which the control policy is learned. Plane space robot systems are used to verify the effectiveness of the presented approach. The simulation results indicate that the proposed method can well achieve the coupling control of the manipulator and the base. Moreover, the space robot can successfully berth the target while suppressing the vibrations of the structure.
What problem does this paper attempt to address?