Reward Design in Reinforcement Learning: How a Single Line Change Can Alter Learning Speed by 100x

Reinforcement learning engineers and researchers often systematically underestimate the importance of the decision to "define ...