AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

CoRe: Combined Rewards with Vision-Language Model Feedback for Preference-Aligned Reinforcement Learning

Hexian Ni, Tao Lu, Yinghao Cai

ICML 2026 · Reinforcement learning › RL from feedback · approach: Theory of RLHF · 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 Reward-Preference Alignment iterations to the formal reward (FRM0 to FRM2) (ablation)Formal Reward Module · MetaWorld Sweep Intosuccess rate (%): 36.0 -> 93.3Improved
more alignment iterations (FRM0 to FRM4) (ablation)Formal Reward Module · MetaWorld Hammersuccess rate (%): 34.4 -> 54.0Improved
add VLM importance weights (KL prior) to residual reward (ablation)Residual Reward Module · MetaWorld Hammersuccess rate (%): 12.7 -> 49.3Improved
add importance weights to residual reward (ablation)RRM · reward vs task progress (10 task avg)Spearman rho: 0.56 -> 0.77Improved
combine FRM with RRM (ablation)RRM alone · MetaWorld Button Presssuccess rate (%): 8.7 -> 100.0Improved
combine RRM with FRM (ablation)FRM4 alone · MetaWorld Hammersuccess rate (%): 54.0 -> 98.0Improved
video-level instead of image-level VLM preference labels (ablation)RRM preference labeling · 10 tasks averagecorrect preference label rate (%): 41.6 -> 51.5Improved
use stronger LLM (GPT-5.4) for FRM (ablation)FRM · MetaWorld Hammersuccess rate (%): 54.0 -> 54.7 (9x cost)No clear effect
use stronger VLM (Gemini 3 Flash) for RRM (ablation)RRM · MetaWorld Hammersuccess rate (%): 49.3 -> 62.6 (6x cost)Improved
replace Gemini 2.0 Flash with open-source Qwen3-VL-8B (ablation)CoRe · MetaWorld Hammersuccess rate (%): 98.0 -> 88.7Hurt
replace Gemini 2.0 Flash with Video-LLaVA-7B (ablation)CoRe · Sweep Into, Hammersuccess rateHurt
add moving red cube visual disturbance (ablation)CoRe · MetaWorld Sweep Intosuccess rate: 97.3% -> 91.7%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