Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Joint Metrics Matter: A Better Standard for Trajectory Forecasting
Erica Weng, Hana Hoshino, Deva Ramanan, Kris Kitani
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 joint (min-over-samples of agent-averaged) reconstruction loss to AgentFormer training | AgentFormer · ETH/UCY avg | min JADE (m): 0.384 -> 0.357 | Improved |
| add joint loss to AgentFormer | AgentFormer · ETH/UCY avg | CR_mean: 0.063 -> 0.054 | Improved |
| add joint loss to AgentFormer | AgentFormer · ETH/UCY avg | min ADE (m): 0.233 -> 0.245 | Hurt |
| add joint loss to View Vertically | View Vertically · ETH/UCY avg | min JADE (m): 0.408 -> 0.392 | Improved |
| add joint loss to View Vertically | View Vertically · ETH/UCY avg | CR_mean: 0.077 -> 0.088 | Hurt |
| add joint loss on dense scenes versus sparse scenes | Joint AgentFormer / Joint View Vertically · ETH/UCY scenes | JADE: 18% and 10% better on zara2; smaller margin on eth, hotel, SDD | Mixed |
| use joint loss only in both AgentFormer training steps instead of marginal+joint in step 1 (ablation) | Joint AgentFormer · ETH/UCY avg | min JADE (m): 0.358 -> 0.386 | Hurt |
| add joint term in step 1 of AgentFormer training (ablation) | Joint AgentFormer · ETH/UCY avg | min JADE (m): 0.365 -> 0.358 | Improved |
| add joint loss in the first training step (ablation) | Joint AgentFormer · ETH/UCY avg | CR_mean: 0.064 -> 0.053 | Improved |
| joint-only loss in both steps (ablation) | AgentFormer · ETH/UCY avg | min ADE (m): 0.233 -> 0.293 | Hurt |
| joint loss reduces collisions by interaction category | Joint AgentFormer · ETH/UCY interaction categories | CR_mean: 0.103 -> 0.084 (group) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AgentFormer — AgentFormer: Agent-Aware Transformers for Socio-Temporal Multi-Agent Forecasting (ICCV 2021) · better on 5 of 7 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
- Collaborative Uncertainty Benefits Multi-Agent Multi-Modal Trajectory Forecasting (TPAMI 2023 · similarity 0.88)
- From Goals, Waypoints & Paths to Long Term Human Trajectory Forecasting (ICCV 2021 · similarity 0.87)
- Collaborative Uncertainty in Multi-Agent Trajectory Forecasting (NeurIPS 2021 · similarity 0.87)
- Evaluation of Trajectory Distribution Predictions with Energy Score (ICML 2024 · similarity 0.85)
- Proper Scoring Rules for Agentic Uncertainty Quantification (NeurIPS 2026 · similarity 0.85)
- Unified Uncertainty-Aware Diffusion for Multi-Agent Trajectory Modeling (CVPR 2025 · similarity 0.85)
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.
