AI Paper Atlas

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

A Dataset for Interactive Vision-Language Navigation with Unknown Command Feasibility

Andrea Burns, Deniz Arsan, Sanjna Agrawal, Ranjitha Kumar, Kate Saenko, Bryan A. Plummer

ECCV 2022 · Vision-language and multimodal › Robot action and navigation · approach: Reasoning-enhanced vision-language navigation · 49 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
concatenate demonstration steps instead of averaging (ablation)CLIP screen-image feasibility MLP · MoTIF feasibilityF1: 44.7 -> 58.2Improved
add app screen image CLIP features to view hierarchy CLIP features (concat) (ablation)feasibility MLP · MoTIF feasibilityF1: 50.9 -> 61.1Improved
replace FastText with CLIP for view hierarchy (concat) (ablation)feasibility MLP · MoTIF feasibilityF1: 43.6 -> 50.9Improved
use Screen2Vec instead of CLIP for view hierarchy (concat) (ablation)feasibility MLP · MoTIF feasibilityF1: 50.9 -> 33.7Hurt
use averaged icon features instead of CLIP screen image (ablation)feasibility MLP · MoTIF feasibility (averaged)F1: 44.7 -> 0.4Hurt
add view hierarchy to CLIP screen image with LSTM aggregation (ablation)feasibility MLP · MoTIF feasibility (LSTM)F1: 42.8 -> 40.9Hurt
add RicoSCA data to MoTIF training (ablation)Seq2Seq · MoTIF complete Action+Ground (app seen)accuracy: 15.9 -> 22.5Improved
add RicoSCA data to MoTIF training (ablation)MOCA · MoTIF complete Action+Ground (app seen)accuracy: 12.3 -> 20.7Improved
add high-level goal to low-level instruction at test time (ablation)Seq2Seq · MoTIF complete Action+Ground (app seen)accuracy: 18.0 -> 22.5Improved
give only high-level goal at test time instead of low-level instruction (ablation)Seq2Act · MoTIF complete Action+Ground (app seen)accuracy: 32.4 -> 7.6Hurt
give only high-level goal instead of both inputs (ablation)MOCA · MoTIF complete Action+Ground (app seen)accuracy: 20.7 -> 8.4Hurt
test on unseen apps instead of seen appsSeq2Seq · MoTIF complete Action+Groundaccuracy: 22.5 -> 17.7Hurt

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