DrivenRewardDefinition

Neural Information Processing Systems 

The design of a reward function often poses a major practical challenge to realworld applications of reinforcement learning.