Efficient reinforcement learning with trials-spanning learning scale for sequential decision-making
Bai Chen, Du Xiuting - · 2011
In this paper, the learning scale is redefined as the integrated scale of learning resources and learning outcomes, based on which two theoretical approaches of extending the scale of learning outcomes are proposed. As an application of the theory, the method of reinforcement learning with trials-spanning learning scale is proposed to combining the spatial and temporal extension of learning scale. The method is applied to the robot path planning problem, which is a classical sequential decision-making problem, in comparison with traditional learning to justify the effectiveness and efficiency of the method.