The End of Reward Engineering: How LLMs Are Redefining Multi-Agent Coordination
This addresses the fundamental problem of manual reward specification for researchers and practitioners in multi-agent systems, representing a novel paradigm shift rather than an incremental improvement.
The paper tackles the challenge of reward engineering in multi-agent reinforcement learning by proposing a shift to using large language models for language-based objective specifications, which can synthesize and adapt rewards from natural language descriptions, as evidenced by methods like EUREKA, CARD, and RLVR.
Reward engineering, the manual specification of reward functions to induce desired agent behavior, remains a fundamental challenge in multi-agent reinforcement learning. This difficulty is amplified by credit assignment ambiguity, environmental non-stationarity, and the combinatorial growth of interaction complexity. We argue that recent advances in large language models (LLMs) point toward a shift from hand-crafted numerical rewards to language-based objective specifications. Prior work has shown that LLMs can synthesize reward functions directly from natural language descriptions (e.g., EUREKA) and adapt reward formulations online with minimal human intervention (e.g., CARD). In parallel, the emerging paradigm of Reinforcement Learning from Verifiable Rewards (RLVR) provides empirical evidence that language-mediated supervision can serve as a viable alternative to traditional reward engineering. We conceptualize this transition along three dimensions: semantic reward specification, dynamic reward adaptation, and improved alignment with human intent, while noting open challenges related to computational overhead, robustness to hallucination, and scalability to large multi-agent systems. We conclude by outlining a research direction in which coordination arises from shared semantic representations rather than explicitly engineered numerical signals.