Research map / Probabilistic and scientific modeling / Molecules and proteins
DSDNet: Deep Structured self-Driving Network
Wenyuan Zeng, Shenlong Wang, Renjie Liao, Yun Chen, Bin Yang, Raquel Urtasun
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 multi-modal future modeling (ablation) | DSDNet prediction · ATG4D prediction ablation | Pred collision (per mille): 4.9 -> 2.4 | Improved |
| add multi-modal future modeling (ablation) | DSDNet prediction · ATG4D prediction ablation | Pred L2 (m): 1.28 -> 1.22 | Improved |
| add pairwise actor interaction modeling (ablation) | DSDNet prediction · ATG4D prediction ablation | Pred collision (per mille): 2.4 -> 0.2 | Improved |
| add pairwise actor interaction modeling (ablation) | DSDNet prediction · ATG4D prediction ablation | Pred L2 (m): 1.22 -> 1.22 | No clear effect |
| condition planner on most likely prediction (ablation) | DSDNet planner · ATG4D planning ablation | Plan collision (%): 0.60 -> 0.45 | Improved |
| use multi-modal predictions instead of most likely in planner (ablation) | DSDNet planner · ATG4D planning ablation | Plan collision (%): 0.45 -> 0.26 | Improved |
| use multi-modal predictions instead of most likely in planner (ablation) | DSDNet planner · ATG4D planning ablation | Lane violation (%): 1.60 -> 1.55 | Improved |
| add prediction to planner (vs none) (ablation) | DSDNet planner · ATG4D planning ablation | Lane violation (%): 1.64 -> 1.55 | Improved |
| replace imitation-learning regression with structured planning cost on predictions | DSDNet · ATG4D planning (3s cumulative) | Collision rate 3s (%): 1.72 -> 0.26 | Improved |
| replace imitation-learning regression with structured planning cost on predictions | DSDNet · ATG4D planning (3s cumulative) | L2 3s (m): 1.92 -> 2.02 | Hurt |
| condition planning on predictions rather than only features (vs NMP) | DSDNet · ATG4D planning (3s cumulative) | Collision rate 3s (%): 0.78 -> 0.26 | Improved |
| share one backbone for detection, prediction and planning | DSDNet · ATG4D detection | Det AP @IoU 0.8: 53.7 -> 55.4 | Improved |
Closest papers by meaning
- SelfD: Self-Learning Large-Scale Driving Policies From the Web (CVPR 2022 · similarity 0.85)
- End-To-End Interpretable Neural Motion Planner (CVPR 2019 · similarity 0.85)
- MotionNet: Joint Perception and Motion Prediction for Autonomous Driving Based on Bird's Eye View Maps (CVPR 2020 · similarity 0.85)
- PnPNet: End-to-End Perception and Prediction With Tracking in the Loop (CVPR 2020 · similarity 0.85)
- SmartPretrain: Model-Agnostic and Dataset-Agnostic Representation Learning for Motion Prediction (ICLR 2025 · similarity 0.84)
- Graph-Based Prediction and Planning Policy Network (GP3Net) for Scalable Self-Driving in Dynamic Environments Using Deep Reinforcement Learning (AAAI 2024 · similarity 0.83)
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.
