Using Human Demonstrations to Improve Reinforcement Learning

摘要

This work introduces Human-Agent Transfer (HAT), an algo-rithm that combines transfer learning, learning from demon-stration and reinforcement learning to achieve rapid learning and high performance in complex domains. Using experi-ments in a simulated robot soccer domain, we show that hu-man demonstrations transferred into a baseline policy for an agent and refined using reinforcement learning significantly improve both learning time and policy performance. Our evaluation compares three algorithmic approaches to incor-porating demonstration rule summaries into transfer learning, and studies the impact of demonstration quality and quantity. Our results show that all three transfer methods lead to statis-tically significant improvement in performance over learning without demonstration.