Research map / Label-efficient and robust learning / Recognition and long-tail
RAP: 3D Rasterization Augmented End-to-End Planning
Lan Feng, Yang Gao, Eloi Zablocki, Quanyi Li, Wuyang Li, Sichao Liu, Matthieu Cord, Alexandre Alahi
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 recovery-oriented perturbed ego trajectories (8.5k) (ablation) | RAP-DINO · NAVSIM v2 two-stage | EPDMS: 32.5 -> 36.9 | Improved |
| add recovery-oriented perturbed ego trajectories (8.5k) (ablation) | RAP-DINO · NAVSIM v1 | PDMS: 92.5 -> 92.5 | No clear effect |
| use colored cuboid faces rather than transparent (ablation) | RAP-ResNet rasterization · NAVSIM navtrain val | MinADE: 0.98 -> 0.91 | Improved |
| add depth decay in rasterization (ablation) | RAP-ResNet rasterization · NAVSIM navtrain val | MinADE: 1.05 -> 0.91 | Improved |
| use black background instead of natural sky-ground (ablation) | RAP-ResNet rasterization · NAVSIM navtrain val | MinADE: 1.33 -> 0.91 | Improved |
| add spatial and global R2R alignment (ablation) | RAP-ResNet · NAVSIM navtrain val (varying real-data fraction) | MinADE | Improved |
| align real features to raster features rather than the reverse or symmetric (ablation) | R2R alignment · NAVSIM navtrain val, 50% real data | MinADE: 1.12 -> 1.02 | Improved |
| scale cross-agent synthetic samples from 1k to 1000k (ablation) | RAP-ResNet · NAVSIM navtrain val | MinADE: y = -0.021 ln(x) + 1.2173 | Improved |
| replace 50% of real data with rasterized samples (with alignment) (ablation) | RAP-ResNet · NAVSIM navtrain val | MinADE | Improved |
| apply RAP to DiffusionDrive | DiffusionDrive · NAVSIM v1 navtest | PDMS: 86.0 -> 89.2 | Improved |
| apply RAP to iPad | iPad · NAVSIM v1 navtest | PDMS: 91.7 -> 92.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DriveTransformer — DriveTransformer: Unified Transformer for Scalable End-to-End Autonomous Driving (ICLR 2025) · 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
- Scenario Dreamer: Vectorized Latent Diffusion for Generating Driving Simulation Environments (CVPR 2025 · similarity 0.86)
- From Representational Complementarity to Dual Systems: Synergizing VLM and Vision-Only Backbones for End-to-End Driving (NeurIPS 2026 · similarity 0.85)
- FutureSightDrive: Thinking Visually with Spatio-Temporal CoT for Autonomous Driving (NeurIPS 2025 · similarity 0.85)
- Explicit Geometric Chain-of-Thought for Vision-Language-Action in Autonomous Driving (NeurIPS 2026 · similarity 0.85)
- RLGF: Reinforcement Learning with Geometric Feedback for Autonomous Driving Video Generation (NeurIPS 2025 · similarity 0.85)
- X-Scene: Large-Scale Driving Scene Generation with High Fidelity and Flexible Controllability (NeurIPS 2025 · 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.
