Simulated Transfer Learning Through Deep Reinforcement Learning

This paper encapsulates the use reinforcement learning on raw images provided by a simulation to produce a partially trained network. Before training is continued, this partially trained network is fed different raw images that are more tightly coupled with a richer representation of the non-simulated environment. The use of transfer learning allows for the model to adjust to this richer representation of the environment and the network eventually exhibits desired behaviours in the real world. This is due to iteratively training the network on gradually more accurate simulated representations.