ConceptReward Shaping / 奖励设计
Reward Shaping
A technique in reinforcement learning concerned with designing and adjusting reward functions. By constructing appropriate reward signals, it guides agents toward desired behaviors. A core challenge in reward shaping is avoiding the reward hacking problem, where agents exploit unintended aspects of the reward function.
Core Facts
Timeline (last 90 days)
Sep 27
Reward shaping refers to adding extra, denser guiding reward signals on top of the original sparse reward to help agents learn faster and more correctly.
Verified75%