A study on designing controller for peg-pushing robot by using reinforcement learning with adaptive state recruitment strategy

摘要

Much attention has been focused on utilizing reinforcement learning (RL) for designing robot controllers. However, as the state spaces of these robots become continuous and high dimensional, it results in time-consuming process. In order to adopt the RL for designing the controllers of such complicated systems, not only adaptability but also computational efficiencies should be taken into account. In this paper, we introduce an adaptive state recruitment strategy, which enables a learning robot to rearrange its state space conveniently according to the task complexity and the progress of the learning.