AI Paper Atlas

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

INSIGHT Bench: Towards Grounded IN-SItu Guidance for Robotic ManipulaTion

Seonho Kim, Junhyeong Hong, Kyungjae Lee, Yoonseon Oh

CVPR 2026 · Vision-language and multimodal › Robot action and navigation · approach: Memory and subgoal-based manipulation

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
fine-tune on data with visual in-situ guides (G-VLA vs NG-VLA) (ablation)pi0 · INSIGHT Bench overallsuccess rate: 12.1% -> 14.1%No clear effect
fine-tune on data with visual in-situ guides (ablation)GR00T N1.5 · INSIGHT Bench overallsuccess rate: 17.2% -> 18.5%No clear effect
fine-tune on data with visual in-situ guides (ablation)SmolVLA · INSIGHT Bench overallsuccess rate: 10.1% -> 7.4%Hurt
add language instruction describing the guide (LI-VLA vs G-VLA) (ablation)GR00T N1.5 · INSIGHT Bench overallsuccess rate: 18.5% -> 24.3%Improved
add language instruction describing the guide (ablation)pi0 · INSIGHT Bench overallsuccess rate: 14.1% -> 18.5%Improved
add visual guides vs none (ablation)GR00T N1.5 · Cabinet (target disambiguation)success rateImproved
add language instruction with position info on top of visual guide (ablation)VLAs · Cabinetsuccess rateNo clear effect
add visual or language guides for directional action (ablation)VLA models · Bottle-Standardsuccess rateNo clear effect
add visual guides on door (ablation)pi0, SmolVLA · Doorsuccess rateHurt
add language instruction (ablation)VLAs · Doorsuccess rateImproved
add language instruction vs visual guide (ablation)pi0 · Bottle-Squeezesuccess rateHurt
add language instruction (ablation)GR00T N1.5 · Bottle-Squeezesuccess rateImproved

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