AI Paper Atlas

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

Learning Affordance Landscapes for Interaction Exploration in 3D Environments

Tushar Nagarajan, Kristen Grauman

NeurIPS 2020 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: Embodied agent learning · 12 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
train exploration to seek novel successful object interactions and learn visual affordance maps onlineIntExp · AI2-iTHOR unseen test environmentsaverage coverage: 16.77 -> 22.15Improved
train exploration to seek novel successful object interactions and learn visual affordance maps onlineIntExp · AI2-iTHOR unseen test environmentsaverage precision: 4.15 -> 8.46Improved
pretrain with interaction exploration before fine-tuningdownstream interaction policies · Retrieve and Storesuccess rate (%): 18.50 -> 27.25; 10.75 -> 27.00Improved
pretrain with interaction exploration before fine-tuningdownstream interaction policies · Wash and Heatsuccess rate (%): 1.50 -> 3.75; 2.25 -> 19.00Improved
use learned affordance maps instead of saliency-only maps (ablation)IntExp affordance model · AI2-iTHOR unseen test environmentsmAP: 13.4 -> 16.3Improved
mark the whole target object rather than a fixed-size point (ablation)IntExp affordance model · AI2-iTHOR unseen test environmentsmAP: 16.3 -> 26.5Improved
use ground-truth affordances instead of affordances learned online (ablation)IntExp affordance model · AI2-iTHOR unseen test environmentsmAP: 26.5 -> 37.3Improved
add an explicit learned affordance map rather than use RGB alone (ablation)IntExp exploration policy · AI2-iTHOR unseen test environmentsinteraction explorationImproved
use the hard count-based reward for each newly discovered interaction instead of softer repeated-interaction reward (ablation)IntExp exploration policy · AI2-iTHOR unseen test environmentsinteraction coverageImproved
use the hard count-based reward for each newly discovered interaction instead of rewarding unique actions regardless of object (ablation)IntExp exploration policy · AI2-iTHOR unseen test environmentsinteraction coverageImproved
interleave affordance training more frequently (ablation)IntExp training schedule · AI2-iTHOR unseen test environmentsmAP: +1.8%Improved
interleave affordance training more frequently (ablation)IntExp training schedule · AI2-iTHOR unseen test environmentsinteraction explorationNo clear effect

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