AI Paper Atlas

Research map / Vision-language and multimodal / Robot action and navigation

Recurrent Reasoning with Vision-Language Models for Estimating Long-Horizon Embodied Task Progress

Yuelin Zhang, Sijie Cheng, Chen Li, Zongzhao Li, Yuxin Huang, Yang Liu, Wenbing Huang

CVPR 2026 · Vision-language and multimodal › Robot action and navigation · approach: Vision-language-action foundation models · 3 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 CoT SFT before RL (instead of direct RL) (ablation)R2VLM · ALFRED progress estimationprogress MAE: 7.21 -> 2.19Improved
continue with RL after SFT (ablation)R2VLM · ALFRED progress estimationtask completion accuracy: 0.961 -> 0.988Improved
continue with RL after SFT (ablation)R2VLM · Ego4D progress estimationprogress MAE: 23.03 -> 19.25Improved
add CoT SFT training to Qwen2.5-VL-7BQwen2.5-VL-7B · ALFRED progress estimationprogress MAE: 27.87 -> 2.77Improved
add improvement reward (ablation)R2VLM RL · ALFRED progress estimationprogress MAE: 2.42 -> 2.19Improved
add MAE reward (ablation)R2VLM RL · ALFRED progress estimationprogress MAE / bin: bin 0.924 -> 0.917; Delta pmae 2.23 -> 2.17Mixed
add bin reward (ablation)R2VLM RL · ALFRED progress estimationprogress MAE: 2.22 -> 2.19Improved
snippet plus historical CoT instead of full video input (ablation)R2VLM · Ego4D progress estimationMAE / inference timeImproved
use R2VLM progress instead of time-based progressSeq2Seq · ALFRED valid seen policy learninggoal-condition completion (%): 8.7 -> 10.9Improved
use R2VLM progress on unseen splitSeq2Seq · ALFRED valid unseen policy learninggoal-condition completion (%): 6.8 -> 7.3No clear effect
use time-based progress supervisionSeq2Seq · ALFRED valid seen policy learninggoal-condition completion (%): 9.4 -> 8.7Hurt
use R2VLM dense reward instead of sparse step rewardSPRINT online RL · EVALINSTRUCT online RL (overall)task completion rate: 0.42 -> 0.50Improved

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