Research map / Multi-agent systems and games / Human-AI and goal-conditioned
From Goals, Waypoints & Paths to Long Term Human Trajectory Forecasting
Karttikeya Mangalam, Yang An, Harshayu Girase, Jitendra Malik
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 Conditioned Waypoint Sampling (CWS) on top of no-CWS, no-TTST Y-net (ablation) | Y-net · SDD long horizon (30 s) | ADE (pixels): 65.00 -> 52.31 | Improved |
| add Conditioned Waypoint Sampling (CWS) on top of no-CWS, no-TTST Y-net (ablation) | Y-net · SDD long horizon (30 s) | FDE (pixels): 86.98 -> 86.98 | No clear effect |
| add Test-Time Sampling Trick (TTST) to Y-net with CWS (ablation) | Y-net · SDD long horizon (30 s) | ADE (pixels): 52.31 -> 47.94 | Improved |
| add Test-Time Sampling Trick (TTST) to Y-net with CWS (ablation) | Y-net · SDD long horizon (30 s) | FDE (pixels): 86.98 -> 66.71 | Improved |
| add Test-Time Sampling Trick (TTST) to Y-net with CWS (ablation) | Y-net · inD long horizon (30 s) | ADE (pixels): 17.77 -> 14.99 | Improved |
| add Test-Time Sampling Trick (TTST) to Y-net with CWS (ablation) | Y-net · inD long horizon (30 s) | FDE (pixels): 28.52 -> 21.13 | Improved |
| increase path multimodality Ka from 1 to 2 (ablation) | Y-net · SDD long horizon (30 s) | ADE (pixels): 47.94 -> 44.94 | Improved |
| increase path multimodality Ka from 1 to 5 (ablation) | Y-net · SDD long horizon (30 s) | ADE (pixels): 47.94 -> 39.49 | Improved |
| increase path multimodality Ka from 1 to 5 (ablation) | Y-net · inD long horizon (30 s) | ADE (pixels): 14.99 -> 12.67 | Improved |
| increase path multimodality Ka from 1 to 5 (ablation) | Y-net · SDD long horizon (30 s) | FDE (pixels): 66.71 -> 66.71 | No clear effect |
| train a separate PECNet per horizon instead of running the 5 s model recurrently | PECNet · SDD long horizon (30 s) | ADE (pixels): 261.27 -> 72.22 | Improved |
| factorized goal and path modeling (Y-net) instead of PECNet | long horizon forecasting · SDD long horizon (30 s) | ADE (pixels): 72.22 -> 47.94 | Improved |
Closest papers by meaning
- Joint Metrics Matter: A Better Standard for Trajectory Forecasting (ICCV 2023 · similarity 0.87)
- Perceiving the Near, Reasoning the Distant: Coherent Long-Horizon Trajectory Prediction for Autonomous Driving (CVPR 2026 · similarity 0.87)
- Multi-modal Knowledge Distillation-based Human Trajectory Forecasting (CVPR 2025 · similarity 0.87)
- HalentNet: Multimodal Trajectory Forecasting with Hallucinative Intents (ICLR 2021 · similarity 0.87)
- TITAN: Future Forecast Using Action Priors (CVPR 2020 · similarity 0.87)
- AutoTraces: Autoregressive Trajectory Forecasting via Multimodal Large Language Models (CVPR 2026 · 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.
