Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Where Will They Go? Predicting Fine-Grained Adversarial Multi-Agent Motion using Conditional Variational Autoencoders
Panna Felsen, Patrick Lucey, Sujoy Ganguly
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add role alignment to VAE (ablation) | VAE · Offense, 4 s | avg distance error (ft): 7.10 -> 6.85 | Improved |
| add history, context and team conditionals to random-aligned VAE (ablation) | CVAE · Offense, 4 s | avg distance error (ft): 7.10 -> 6.76 | Improved |
| add a single conditional (history, context, team or player) with random alignment (ablation) | CVAE · Offense, 4 s | avg distance error (ft): 7.10 -> 6.90 | No clear effect |
| add role alignment to the CVAE with history, context and team (ablation) | CVAE · Offense, 4 s | avg distance error (ft): 6.76 -> 5.80 | Improved |
| use role alignment instead of canonical position alignment (ablation) | CVAE history+context+team · Offense, 4 s | avg distance error (ft): 6.09 -> 5.80 | Improved |
| add team identity conditional under role alignment (ablation) | CVAE role · Offense, 4 s | avg distance error (ft): 5.81 -> 5.80 | No clear effect |
| use player identity instead of team identity under role alignment (ablation) | CVAE role · Offense, 4 s | avg distance error (ft): 5.80 -> 5.96 | Hurt |
| add history and context conditionals to random-aligned CVAE with player identity (ablation) | CVAE random · Offense, 4 s | avg distance error (ft): 7.02 -> 6.64 | Improved |
| replace Social LSTM with CVAE-role-team | mix 5v5 prediction · Basketball 4 s horizon | avg distance error (ft): 17.95 -> 5.74 | Improved |
| increase observed history length (1 s to 4 s) (ablation) | CVAE-role-team · Basketball offense | avg distance error (ft): 4 s history barely optimal | Improved |
| increase prediction horizon from 1 s to 8 s (ablation) | CVAE-role-team offense · Basketball offense | avg distance error (ft): 2.35 -> 8.07 | Hurt |
| increase predicted players from 1v1 to 5v5 (ablation) | CVAE-role-team · Basketball mixture 4 s | avg distance error (ft): 4.19 -> 5.74 | Hurt |
Closest papers by meaning
- Semi-Supervised Generative Models for Multiagent Trajectories (NeurIPS 2022 · similarity 0.89)
- Stochastic Prediction of Multi-Agent Interactions from Partial Observations (ICLR 2019 · similarity 0.87)
- Latent Variable Sequential Set Transformers for Joint Multi-Agent Motion Prediction (ICLR 2022 · similarity 0.86)
- An Uncertain Future: Forecasting from Static Images Using Variational Autoencoders (ECCV 2016 · similarity 0.86)
- Quantifying Uncertainty in Motion Prediction with Variational Bayesian Mixture (CVPR 2024 · similarity 0.84)
- PlaySlot: Learning Inverse Latent Dynamics for Controllable Object-Centric Video Prediction and Planning (ICML 2025 · similarity 0.84)
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.
