AI Paper Atlas

Research map / Multi-agent systems and games / Human-AI and goal-conditioned

Auxiliary Tasks and Exploration Enable ObjectGoal Navigation

Joel Ye, Dhruv Batra, Abhishek Das, Erik Wijmans

ICCV 2021 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: Vision-language navigation agents · 164 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 Semantic Goal Exists (SGE) input feature (ablation)4-Action agent · MP3D VALsuccess %: 20.3 -> 34.4Improved
add Semantic Goal Exists (SGE) input feature (ablation)4-Action agent · MP3D VALSPL %: 4.14 -> 9.58Improved
add Coverage Prediction auxiliary task (ablation)4-Action agent · MP3D VALsuccess %: 32.4 -> 34.4Improved
add Generalized Inverse Dynamics auxiliary task (ablation)4-Action agent · MP3D VALsuccess %: 33.7 -> 34.4Improved
add Action Distribution Prediction auxiliary task (ablation)4-Action agent · MP3D VALsuccess %: 34.1 -> 34.4No clear effect
use tethered policy instead of direct sparse-reward finetuning (ablation)6-Action agent · MP3D VALsuccess %: 23.1 -> 26.6Improved
use tethered policy instead of direct sparse-reward finetuning (ablation)6-Action agent · MP3D VALSPL %: 8.43 -> 9.79Improved
add tethered sparse-reward policy6-Action agent · MP3D VALSPL %: SPL 7.60 -> 9.79, success 30.8 -> 26.6Mixed
finetune with RedNet segmentation4-Action agent · MP3D VALsuccess %: 34.4 -> 33.1Hurt
finetune with RedNet segmentation6-Action agent · MP3D VALsuccess %: 30.8 -> 34.6Improved
add LOOK_UP and LOOK_DOWN actions (6-action vs 4-action)agent with GT segmentation · MP3D VAL subset (300 episodes, GT segm.)success %: 43.3 -> 58.0Improved
replace GT segmentation with RedNet predicted segmentation6-Action agent · MP3D VAL subset (300 episodes)success %: 58.0 -> 30.8Hurt

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