AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

GTR: Guided Thought Reinforcement Prevents Thought Collapse in RL-based VLM Agent Training

Tong Wei, Yijun Yang, Junliang Xing, Yuanchun Shi, Zongqing Lu, Deheng Ye

ICCV 2025 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards · 13 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 process guidance via VLM corrector (GTR) over outcome-only RLRL4VLM (PPO) · gym_cards EZPointssuccess rate (%): 48.0 -> 94.5Improved
add GTR over RL4VLMLLaVA-v1.6-mistral-7B · gym_cards Numberlinesuccess rate (%): 90.5 -> 100.0Improved
add GTR over RL4VLMLLaVA-v1.6-mistral-7B · gym_cards Blackjacksuccess rate (%): 40.1 -> 41.3Improved
add GTR over RL4VLMLLaVA-v1.6-mistral-7B · gym_cards EZPointsepisode return: 4.19 -> 9.43Improved
add GTR over RL4VLM, no text observationLLaVA-v1.6-mistral-7B · ALFWorld (without text)success rate (%): 5.4 -> 17.8Improved
add text description of observation to RL4VLMRL4VLM · ALFWorldsuccess rate (%): 5.4 -> 21.7Improved
add text description of observation to GTRGTR · ALFWorldsuccess rate (%): 17.8 -> 21.0Improved
use GPT-4o instead of Qwen2.5-VL-72B as corrector (ablation)GTR · Points24performance (%): 6.5 -> 17.5Improved
use GPT-4o instead of Qwen2.5-VL-7B as corrector (ablation)GTR · Points24performance: N/A -> 17.5Improved
keep thought-cloning loss throughout training instead of cosine annealing it (ablation)GTR · Points24training curveImproved
add tool function calling to GPT-4o corrector (ablation)GTR · Points24training curveImproved
add DAgger aggregation of thought dataset (ablation)GTR · Points24training curveImproved

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