Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Social ODE: Multi-agent Trajectory Forecasting with Neural Ordinary Differential Equations
Song Wen, Hao Wang, Dimitris N. Metaxas
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 spatio-temporal Transformer encoder (vs Latent ODE encoder) with the interaction decoder (ablation) | Social ODE · inD curve (4s) | ADE: 3.2147 -> 3.1417 | Improved |
| add spatio-temporal Transformer encoder (vs Social LSTM encoder) (ablation) | Social ODE · highD curve (4s) | ADE: 3.3256 -> 3.1509 | Improved |
| add distance, interaction intensity and aggressiveness decoder (vs plain Neural ODE) (ablation) | Social ODE · rounD curve (4s) | ADE: 2.6227 -> 2.4738 | Improved |
| add distance, interaction intensity and aggressiveness terms (vs social pooling ODE) (ablation) | Social ODE · highD curve (4s) | ADE: 3.2740 -> 3.1509 | Improved |
| add distance, interaction intensity and aggressiveness terms (vs social pooling ODE) (ablation) | Social ODE · inD straight (4s) | ADE: 0.7857 -> 0.7728 | Improved |
| add Social ODE interaction modeling (vs Social LSTM) | trajectory forecaster · Sudden static obstacle | collision rate: 28.6% -> 8.8% | Improved |
| add Social ODE interaction modeling (vs DenseTNT) | trajectory forecaster · Sudden moving obstacle | collision rate: 32.6% -> 12.8% | Improved |
| increase attractor coefficient lambda from 0 to 10 | Social ODE · Target reaching | reaching rate: 4.8% -> 87.8% | Improved |
| Social ODE versus AgentFormer on forecast accuracy | Social ODE · inD straight (4s) | ADE: 0.7604 vs 0.7728 | Hurt |
| Social ODE versus AgentFormer on curved trajectories | Social ODE · rounD curve (4s) | ADE: 2.4748 -> 2.4738 | No clear effect |
| Social ODE versus AgentFormer on 8s straight rounD | Social ODE · rounD straight (8s) | ADE: 2.5337 vs 2.6447 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- AgentFormer — AgentFormer: Agent-Aware Transformers for Socio-Temporal Multi-Agent Forecasting (ICCV 2021) · better on 3 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- AgentFormer: Agent-Aware Transformers for Socio-Temporal Multi-Agent Forecasting (ICCV 2021 · similarity 0.91)
- SMEMO: Social Memory for Trajectory Forecasting (TPAMI 2024 · similarity 0.89)
- Trajectron++: Dynamically-Feasible Trajectory Forecasting With Heterogeneous Data (ECCV 2020 · similarity 0.89)
- Social-Mamba: Socially-Aware Trajectory Forecasting with State-Space Models (ECCV 2026 · similarity 0.88)
- PGODE: Towards High-quality System Dynamics Modeling (ICML 2024 · similarity 0.88)
- GRIN: Generative Relation and Intention Network for Multi-agent Trajectory Prediction (NeurIPS 2021 · similarity 0.87)
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.
