GD$^2$PO: Mitigating Multi-Reward Conflicts via Group-Dynamic reward-Decoupled Policy Optimization

2026-06-15

AI Summary

This work addresses the challenge in multi-reward reinforcement learning where conflicting reward signals often lead to mutual cancellation of advantage estimates, thereby degrading policy optimization efficiency. To mitigate this issue, the authors propose a conflict-aware dynamic filtering mechanism that automatically masks trajectories exhibiting high inter-reward disagreement, coupled with a query-level adaptive reweighting strategy that dynamically modulates the update intensity of each reward dimension. Built upon a group-decoupled optimization framework, the method integrates advantage masking with dynamic sampling to effectively alleviate reward conflicts. Empirical evaluations on tasks such as tool-augmented reasoning and human preference alignment demonstrate substantial improvements over existing baselines, confirming the approach’s efficacy and robustness in multi-reward settings.

Abstract

As LLMs advance, post-training reinforcement learning (RL) increasingly relies on multi-dimensional rewards to cultivate comprehensive capabilities. This shift demands new algorithms capable of optimizing diverse and potentially competing objectives simultaneously. To address this, existing methods such as Group reward-Decoupled Policy Optimization (GDPO) decompose the overall score into independent reward groups, then compute the RL loss separately within each group. However, this strategy still encounters multi-reward conflicts: a single rollout can yield positive advantages on certain reward dimensions but negative ones on others, causing opposing signals to cancel each other out during aggregation, further hindering RL training efficiency. Inspired by Dynamic sAmpling Policy Optimization (DAPO), which improves RL training efficiency by filtering out ineffective rollouts with near-zero advantages, we propose Group-Dynamic reward-Decoupled Policy Optimization (GD$^2$PO). Specifically, GD$^2$PO employs a conflict-aware filtering mechanism to mask out rollouts suffering from severe reward-wise disagreement. By preventing conflicting signals from canceling each other out, this masking strategy preserves and enhances the magnitude of effective RL advantages, thereby significantly accelerating learning efficiency. Furthermore, we introduce query-level reweighting to dynamically adjust the update intensity of each query based on its overall reward consensus. Experiments on various multi-reward scenarios, including tool calling and human preference alignment, demonstrate that GD$^2$PO consistently and significantly outperforms existing baselines. The code is available at https://github.com/Qwen-Applications/GD2PO.

Problem

• multi-reward conflicts
• reinforcement learning
• reward decomposition
• policy optimization
• conflicting signals

Innovation

• multi-reward conflict
• reward-decoupled policy optimization
• conflict-aware filtering
• dynamic reweighting
• reinforcement learning for LLMs