Multiple Model-Based Reinforcement Learning

  title={Multiple Model-Based Reinforcement Learning},
  author={Kenji Doya and Kazuyuki Samejima and Ken-ichi Katagiri and Mitsuo Kawato},
  journal={Neural Computation},
We propose a modular reinforcement learning architecture for nonlinear, nonstationary control tasks, which we call multiple model-based reinforcement learning (MMRL). The basic idea is to decompose a complex task into multiple domains in space and time based on the predictability of the environmental dynamics. The system is composed of multiple modules… CONTINUE READING