Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Reverberation: Learning the Latencies Before Forecasting Trajectories
Conghao Wong, Ziqian Zou, Beihao Xia, Xinge You
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 |
|---|---|---|---|
| increase generating kernel size Kg from 20 to 40 in the v_soc variant (ablation) | social reverberation kernel R_soc · ETH-UCY rank analysis (8,12,0.4), 10,000 samples | full-rank rate (%): 99.92% -> 100.00% | Improved |
| extend social branch mapping from T_h N_theta -> T_f to T_h N_theta -> T_f N_theta (v_soc) (ablation) | Rev social branch · ETH-UCY | ADE/FDE: about +-1% ADE and FDE | No clear effect |
| lengthen the sequence from (8,12) to (16,24) steps | non-interactive self-attention matrix W_non · ETH-UCY rank analysis | full-rank rate (%): 99.44% -> 80.05% | Hurt |
| replace self-attention weights with reverberation kernel in the social branch | social branch rank · ETH-UCY rank analysis (8,12,0.4), 10,000 samples | full-rank rate (%): 0.00% -> 100.00% | Improved |
| lengthen the sequence from (8,12) to (16,24) steps | non-interactive reverberation kernel R_non · ETH-UCY rank analysis | full-rank rate (%): 100.00% -> 100.00% | No clear effect |
| reverse the columns of kernel R (ablation) | Rev predictions · ETH-UCY | predicted trajectories | Hurt |
| exchange first and last half of columns of kernel R (ablation) | Rev predictions · ETH-UCY | predicted trajectories | Hurt |
| reset first half of R columns to zero (ablation) | Rev predictions · ETH-UCY | predicted trajectories | Hurt |
| randomize last column of R with uniform noise (ablation) | Rev predictions · ETH-UCY | predicted trajectories | Hurt |
Closest papers by meaning
- Multiple Futures Prediction (NeurIPS 2019 · similarity 0.86)
- Adaptive Human Trajectory Prediction via Latent Corridors (ECCV 2024 · similarity 0.86)
- Resonance: Learning to Predict Social-Aware Pedestrian Trajectories as Co-Vibrations (ICCV 2025 · similarity 0.86)
- AutoTraces: Autoregressive Trajectory Forecasting via Multimodal Large Language Models (CVPR 2026 · similarity 0.85)
- Towards Understanding Future: Consistency Guided Probabilistic Modeling for Action Anticipation (AAAI 2024 · similarity 0.85)
- Latent Plan Transformer for Trajectory Abstraction: Planning as Latent Space Inference (NeurIPS 2024 · 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.
