ARIA: Training Language Agents with Intention-Driven Reward Aggregation

TL;DR

ARIA introduces a novel reward aggregation method in intention space that reduces reward variance and enhances training efficiency for language agents in complex, open-ended environments, leading to notable performance improvements.

Contribution

The paper presents ARIA, a new approach that projects language actions into an intention space to improve reward signal density and reinforcement learning effectiveness.

Findings

Reduces policy gradient variance significantly.

Achieves an average of 9.95% performance improvement.

Outperforms baseline RL methods across tasks.

Abstract

Large language models (LLMs) have enabled agents to perform complex reasoning and decision-making through free-form language interactions. However, in open-ended language action environments (e.g., negotiation or question-asking games), the action space can be formulated as a joint distribution over tokens, resulting in an exponentially large action space. Sampling actions in such a space can lead to extreme reward sparsity, which brings large reward variance, hindering effective reinforcement learning (RL). To address this, we propose ARIA, a method that Aggregates Rewards in Intention space to enable efficient and effective language Agents training. ARIA aims to project natural language actions from the high-dimensional joint token distribution space into a low-dimensional intention space, where semantically similar actions are clustered and assigned shared rewards. This intention-aware reward aggregation reduces reward variance by densifying reward signals, fostering better policy optimization. Extensive experiments demonstrate that ARIA not only significantly reduces policy gradient variance, but also delivers substantial performance gains of an average of 9.95% across four downstream tasks, consistently outperforming offline and online RL baselines.