Reward-Zero: Language Embedding Driven Implicit Reward Mechanisms for Reinforcement Learning
arXiv:2603.09331v1 Announce Type: new Abstract: We introduce Reward-Zero, a general-purpose implicit reward mechanism that transforms natural-language task descriptions into dense, semantically grounded progress signals for …
Heng Zhang, Haddy Alchaer, Arash Ajoudani, Yu She
13 views