AI Paper Atlas

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

Pathdreamer: A World Model for Indoor Navigation

Jing Yu Koh, Honglak Lee, Yinfei Yang, Jason Baldridge, Peter Anderson

ICCV 2021 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: Vision-language navigation agents · 84 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
replace repeated-panorama look-ahead with Pathdreamer predictions (3 planning steps)VLN agent with compatibility model · R2R Val-Unseen VLNSR: 40.6 -> 50.4Improved
replace repeated-panorama look-ahead with Pathdreamer predictions (1 planning step)VLN agent with compatibility model · R2R Val-Unseen VLNSR: 35.7 -> 39.9Improved
replace blank-panorama look-ahead with Pathdreamer predictions (2 planning steps)VLN agent with compatibility model · R2R Val-Unseen VLNSR: 40.0 -> 46.5Improved
use ground-truth look-ahead observations instead of Pathdreamer (3 planning steps)VLN agent with compatibility model · R2R Val-Unseen VLNSR: 50.4 -> 59.3Improved
increase planning horizon from 1 to 3 steps with ground-truth observationsVLN agent · R2R Val-Unseen VLNSR: 44.6 -> 59.3Improved
increase planning horizon from 1 to 3 steps with repeated panoramaVLN agent · R2R Val-Unseen VLNSR: 35.7 -> 40.6Improved
add semantic and depth inputs to the Image Generator (ablation)Image Generator · Val-Unseen RGB, 1-6 steps, 1 contextFID: 90.8 -> 39.9Improved
add RGB context and guidance image to the Image Generator (vs SPADE) (ablation)Image Generator · Val-Unseen RGB, 1-6 steps, 1 contextFID: 50.3 -> 39.9Improved
add RGB context and guidance image to the Image Generator (vs SPADE) (ablation)Image Generator · Val-Seen RGB, 1-6 steps, 1 contextFID: 24.9 -> 31.3Hurt
use ground-truth semantics instead of Structure Generator predictionsImage Generator · Val-Unseen RGB, 1-6 steps, 1 contextFID: 70.4 -> 39.9Improved
use the learned Structure Generator instead of nearest-neighbor hole fillingStructure Generator · Val-Unseen semantics, 1-6 steps, 1 contextmIoU: 30.6 -> 50.8Improved
use recurrent training instead of teacher forcing (ablation)Structure Generator · Val-Seen semantics, 1-6 steps, 1 contextmIoU: 59.2 -> 65.9Improved

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