Who Deserves the Reward? SHARP: Shapley Credit-based Optimization for Multi-Agent System

TL;DR

SHARP introduces a Shapley-based credit assignment framework for multi-agent reinforcement learning, improving training stability and performance by accurately attributing individual contributions in complex multi-agent systems.

Contribution

The paper proposes SHARP, a novel Shapley-based hierarchical attribution method that enhances credit assignment and training stability in multi-agent reinforcement learning.

Findings

SHARP outperforms state-of-the-art baselines by 23.66% and 14.05% in benchmark tasks.

The framework stabilizes training through normalized agent-specific advantages.

Extensive experiments validate SHARP's effectiveness across real-world benchmarks.

Abstract

Integrating Large Language Models (LLMs) with external tools via multi-agent systems offers a promising new paradigm for decomposing and solving complex problems. However, training these systems remains notoriously difficult due to the credit assignment challenge, as it is often unclear which specific functional agent is responsible for the success or failure of decision trajectories. Existing methods typically rely on sparse or globally broadcast rewards, failing to capture individual contributions and leading to inefficient reinforcement learning. To address these limitations, we introduce the Shapley-based Hierarchical Attribution for Reinforcement Policy (SHARP), a novel framework for optimizing multi-agent reinforcement learning via precise credit attribution. SHARP effectively stabilizes training by normalizing agent-specific advantages across trajectory groups, primarily through a decomposed reward mechanism comprising a global broadcast-accuracy reward, a Shapley-based marginal-credit reward for each agent, and a tool-process reward to improve execution efficiency. Extensive experiments across various real-world benchmarks demonstrate that SHARP significantly outperforms recent state-of-the-art baselines, achieving average match improvements of 23.66% and 14.05% over single-agent and multi-agent approaches, respectively.