AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

CRPO: Character-centric Group Relative Policy Optimization for Role-aware Reasoning in Role-playing Agents

Yihong Tang, Kehai Chen, Liang Yue, Benyou Wang, Min Zhang

ICML 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add dual-stream advantage (ablation)CRPO (Qwen3-8B) · CharacterBenchaverage score: 3.811 -> 4.043Improved
add entropy-aware adaptive exploitation (ablation)CRPO (Qwen3-8B) · CharacterBenchaverage score: 3.962 -> 4.043Improved
add contrastive anchor sampling (ablation)CRPO (Qwen3-8B) · CharacterBenchaverage score: 3.942 -> 4.043Improved
add contrastive anchor sampling (ablation)CRPO (Qwen3-8B) · CharacterBench Engagementscore: 3.025 -> 3.130Improved
add dual-stream advantage (ablation)CRPO (Qwen3-8B) · CharacterBench Attribute Consistency (Bot)score: 4.294 -> 4.994Improved
add entropy-aware adaptive exploitation (ablation)CRPO (Qwen3-8B) · CharacterBench Fact Accuracyscore: 2.553 -> 2.581Improved
use CRPO instead of GRPOQwen3-8B · CharacterBenchaverage score: 3.829 -> 4.043Improved
use CRPO instead of GRPOLlama-3.2-3B-Instruct · CharacterBenchaverage score: 3.781 -> 3.832Improved
add SFT on role-play dataQwen3-8B · CharacterBenchaverage score: 3.701 -> 3.673No clear effect
train with CRPOQwen3-8B · SocialBench (unseen)average score: 0.769 -> 0.802Improved
train with CRPOQwen3-8B · SocialBench SUscore: 0.230 -> 0.350Improved
increase number of contrastive anchors beyond one (ablation)CRPO · CharacterBenchaverage scoreNo clear effect

Later papers that compared with this method

1 later accepted paper reports CRPO as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy