Self-Augmenting Strategy for Reinforcement Learning

Training the agent to interact with environment intelligently is one of the core problems in reinforcement learning. In this paper, we propose a Self-Augmenting strategy for reinforcement learning to accelerate the learning process and agent performance by imitating and augmenting practical human experience. Instead of exploring randomly at the beginning as Deep Q Learning algorithms do, our strategy uses a short series of expert experience of human as augmenting guide in the training for the agent. Because of the imitation from the similar states of the augmented human experience, the agent trained with our strategy scores higher and converges faster than the original Deep Q Learning method does.