Research map / Multi-agent systems and games / Human-AI and goal-conditioned
RAIN: Reinforced Hybrid Attention Inference Network for Motion Forecasting
Jiachen Li, Fan Yang, Hengbo Ma, Srikanth Malla, Masayoshi Tomizuka, Chiho Choi
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 RL-based hard attention (hybrid) instead of soft attention only (ablation) | RAIN · nuScenes vehicles | miss rate @2.0m (%): 21.0 -> 11.3 | Improved |
| add RL-based hard attention (hybrid) instead of soft attention only (ablation) | RAIN · nuScenes pedestrians | miss rate @1.0m (%): 13.2 -> 11.7 | Improved |
| use RL-based hard attention instead of ELBO-based module (ablation) | RAIN · nuScenes vehicles | miss rate @2.0m (%): 16.1 -> 11.3 | Improved |
| use RL-based hard attention instead of ELBO-based module (ablation) | RAIN · Mixed particle system relation recognition | accuracy (%): 89.31 -> 93.52 | Improved |
| use dynamic (recurrent) hard attention instead of static (ablation) | RAIN · nuScenes vehicles | miss rate @2.0m (%): 11.3 -> 9.5 | Improved |
| use dynamic (recurrent) hard attention instead of static (ablation) | RAIN · nuScenes pedestrians | miss rate @1.0m (%): 11.7 -> 9.8 | Improved |
| use dynamic (recurrent) hard attention instead of static (ablation) | RAIN · nuScenes vehicles | minFDE20 at 4.0s (m): 1.17 -> 1.12 | Improved |
| add vehicle kinematics model to motion generator | RAIN (full+soft) · nuScenes vehicles | minFDE20 at 4.0s (m): 1.27 -> 1.28 | No clear effect |
| add hard attention over history soft attention only (HisRepItself) | RAIN · Human3.6M Walking | MAE 1k ms: 0.64 -> 0.60 | Improved |
| add hard attention over history soft attention only (HisRepItself) | RAIN · Human3.6M Discussion | MAE 400 ms: 0.87 -> 0.91 | Hurt |
| use RAIN hybrid attention instead of DNRI | RAIN · Mixed particle system relation recognition | accuracy (%): 87.54 -> 93.52 | Improved |
| use RAIN instead of fully supervised relation labels | RAIN · Mixed particle system relation recognition | accuracy (%): 97.48 -> 93.52 | Hurt |
Closest papers by meaning
- DeMo: Decoupling Motion Forecasting into Directional Intentions and Dynamic States (NeurIPS 2024 · similarity 0.87)
- TPCN: Temporal Point Cloud Networks for Motion Forecasting (CVPR 2021 · similarity 0.87)
- SHARP: Short-Window Streaming for Accurate and Robust Prediction in Motion Forecasting (CVPR 2026 · similarity 0.87)
- HPNet: Dynamic Trajectory Forecasting with Historical Prediction Attention (CVPR 2024 · similarity 0.86)
- Motion Forecasting in Continuous Driving (NeurIPS 2024 · similarity 0.86)
- Trajectory Mamba: Efficient Attention-Mamba Forecasting Model Based on Selective SSM (CVPR 2025 · similarity 0.86)
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.
