AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

RLVMR: Reinforcement Learning with Verifiable Meta-Reasoning Rewards for Robust Long-Horizon Agents

Zijing Zhang, Ziyang Chen, Mingxiao Li, Zhaopeng Tu, Xiaolong Li

ICLR 2026 · Reinforcement learning › Visual and representation RL · approach: Probabilistic and autonomous RL

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 verifiable meta-reasoning rewards to GRPO-style RLQwen2.5-1.5B on ALFWorld · ALFWorld L2success rate (%): 29.7 -> 56.3Improved
use RLVMR instead of GiGPOQwen2.5-7B on ALFWorld · ALFWorld L2success rate (%): 67.2 -> 83.6Improved
use RLVMR instead of GiGPO on ScienceWorld seen splitQwen2.5-7B · ScienceWorld L0success rate (%): 53.4 -> 67.2Improved
remove outcome-based trajectory advantage (ablation)RLVMR 1.5B · ALFWorld L2success rate (%): 12.5 -> 56.3Improved
add meta-reasoning advantage signal (ablation)RLVMR 1.5B · ALFWorld L2success rate (%): 45.3 -> 56.3Improved
add cold-start SFT on 200 trajectories (ablation)RLVMR 1.5B · ALFWorld L2success rate (%): 40.6 -> 56.3Improved
use RLVMR instead of GRPO to reduce repetitive actionsQwen2.5-7B · ALFWorld L0repetitive action rate (%): 21.5 -> 2.3Improved
use RLVMR instead of GRPO for episode lengthQwen2.5-7B · ALFWorld L2avg actions: 21.7 -> 15.4Improved
scale model from 1.5B to 7B under GRPOGRPO · ALFWorld L2repetitive action rate / success: success 29.7 -> 52.3; repetitive 27.1 -> 31.2Mixed
SFT on expert trajectories vs GRPO on unseen L2Qwen2.5-7B · ALFWorld L2success rate (%): 52.3 (GRPO) -> 37.5 (SFT)Hurt

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