Research map / Multi-agent systems and games / Human-AI and goal-conditioned
CAR-Net: Clairvoyant Attentive Recurrent Network
Amir Sadeghian, Ferdinand Legros, Maxime Voisin, Ricky Vesel, Alexandre Alahi, Silvio Savarese
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 single-source and multi-source attention outputs together (ablation) | SS-LSTM · Car-Racing (optimal trajectories) | ML2: 4.64 -> 3.58 | Improved |
| add single-source and multi-source attention outputs together (ablation) | MS-LSTM · Car-Racing (optimal trajectories) | ML2: 5.30 -> 3.58 | Improved |
| combine both attentions (ablation) | SS-LSTM · Formula 1 (optimal trajectories) | ML2: 14.7 -> 13.3 | Improved |
| combine both attentions (ablation) | MS-LSTM · SDD (12 future from 8 past positions) | ML2: 27.38 -> 25.72 | Improved |
| add attention over image features (ablation) | I-LSTM · Formula 1 (optimal trajectories) | ML2: 20.8 -> 14.7 | Improved |
| add whole-image features without attention (ablation) | T-LSTM · Formula 1 (optimal trajectories) | ML2: 21.2 -> 20.8 | Improved |
| add scene image context with attention (ablation) | T-LSTM · Car-Racing (middle trajectories) | ML2: 10.4 -> 5.0 | Improved |
| use single-source instead of multi-source attention on large racing track images (ablation) | MS-LSTM · Formula 1 (optimal trajectories) | ML2: 18.9 -> 14.7 | Improved |
| use single-source instead of multi-source attention on SDD (ablation) | MS-LSTM · SDD (12 future from 8 past positions) | FL2: 52.69 -> 63.27 | Hurt |
| use CAR-Net scene context on semantically complex scenes (ablation) | T-LSTM · SDD semantically complex scenes | ML2: 31.31 -> 24.32 | Improved |
| use CAR-Net scene context on semantically simple scenes (ablation) | T-LSTM · SDD semantically simple scenes | ML2: 30.48 -> 30.92 | No clear effect |
| add scene attention to trajectory LSTM | T-LSTM · SDD (12 future from 8 past positions) | ML2: 31.96 -> 25.72 | Improved |
Later papers that compared with this method
1 later accepted paper reports CAR-Net as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Trajectron++: Dynamically-Feasible Trajectory Forecasting With Heterogeneous Data (ECCV 2020 · better on 1 of 1 shared result)
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
- SMART: Simultaneous Multi-Agent Recurrent Trajectory Prediction (ECCV 2020 · similarity 0.87)
- VectorNet: Encoding HD Maps and Agent Dynamics From Vectorized Representation (CVPR 2020 · similarity 0.87)
- TITAN: Future Forecast Using Action Priors (CVPR 2020 · 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)
- Think Before You Drive: World Model-Inspired Multimodal Grounding (CVPR 2026 · 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.
