AI Paper Atlas

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

Social NCE: Contrastive Learning of Socially-Aware Motion Representations

Yuejiang Liu, Qi Yan, Alexandre Alahi

ICCV 2021 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: Interactive multi-agent trajectory prediction · 123 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 Social-NCE lossSocial-STGCNN · ETH/UCY averagecollision rate (%): 4.70 -> 2.96Improved
add Social-NCE lossTrajectron++ · ETH/UCY averagecollision rate (%): 1.37 -> 0.93Improved
add Social-NCE lossSocial-STGCNN · ETH/UCY averagetop-20 FDE: 0.765 -> 0.756No clear effect
add Social-NCE lossTrajectron++ · ETH/UCY averagetop-20 FDE: 0.403 -> 0.398No clear effect
add Social-NCE lossSocial-LSTM · Trajnet++ overallcollision rate (%): 6.44 -> 5.31Improved
add Social-NCE lossDirectional-LSTM · Trajnet++ overallcollision rate (%): 5.49 -> 4.59Improved
add Social-NCE loss with social negativesbehavioral cloning · crowd navigation, 5k demoscollision rate (%): 11.11 -> 3.40Improved
replace social negatives with random negative sampling (ablation)Social-NCE · crowd navigation, 5k demoscollision rate (%): 3.40 -> 18.60 (vanilla 11.11)Hurt
add Social-NCE lossRainbow DQN offline · crowd navigation, 10% of datanormalized score (%): 14.4% -> 69.0%Improved
add random-negative contrastive lossRainbow DQN offline · crowd navigation, 10% of datanormalized score (%): 14.4% -> 7.0%Hurt
add Social-NCE lossRainbow DQN online · crowd navigationepisodes to reach reward 0.6: more than 4000 -> less than 2000Improved
use multiple sampling horizons 1-4 instead of single horizon 1 (ablation)Social-NCE · behavioral cloningcollision rate (%): 11.24 -> 3.40Improved

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