Research map / 3D vision / 3D generation
Model-Based Imitation Learning for Urban Driving
Anthony Hu, Gianluca Corrado, Nicolas Griffiths, Zachary Murez, Corina Gurau, Hudson Yeo, Alex Kendall, Roberto Cipolla, Jamie Shotton
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 3D lifting of image features to BeV (ablation) | MILE · CARLA Town05 new town | cumulative reward: 4564 -> 7621 | Improved |
| add 3D lifting of image features to BeV (ablation) | single-frame model · CARLA Town05 new town | cumulative reward: 1878 -> 6630 | Improved |
| add 3D lifting of image features (ablation) | MILE · CARLA Town05 new town | driving score: 63.0 -> 61.1 | Mixed |
| add KL prior/posterior matching (ablation) | MILE · CARLA Town05 new town | cumulative reward: 6084 -> 7621 | Improved |
| add KL prior/posterior matching (ablation) | MILE · CARLA Town05 new town | driving score: 63.3 -> 61.1 | No clear effect |
| add BeV segmentation supervision (ablation) | MILE · CARLA Town05 new town | driving score: 55.0 -> 61.1 | Improved |
| add recurrent memory (sequence model) over single frame (ablation) | MILE · CARLA Town05 new town | cumulative reward: 6630 -> 7621 | Improved |
| deploy fully recurrent instead of resetting state (ablation) | MILE · CARLA Town05 new town | cumulative reward: 7621 -> 7532 | No clear effect |
| add Gaussian noise to past latent state at deployment (ablation) | MILE fully recurrent · CARLA Town05 new town | cumulative reward: 7532 -> 3603 | Hurt |
| use spatial latent state instead of 512-d vector (ablation) | MILE reset state · CARLA Town05 new town | cumulative reward: 7621 -> 7465, 6407, 5637 | Hurt |
| increase image resolution from 75x120 to 600x960 (ablation) | MILE · CARLA Town05 new town | driving score: 20.9 -> 61.1 | Improved |
| increase image resolution from 75x120 to 600x960 (ablation) | MILE · CARLA Town05 new town | red lights run per km: 3.07 -> 0.13 | Improved |
Closest papers by meaning
- Keyframe-Focused Visual Imitation Learning (ICML 2021 · similarity 0.89)
- DrivingGen: A Comprehensive Benchmark for Generative Video World Models in Autonomous Driving (ICLR 2026 · similarity 0.89)
- LASIL: Learner-Aware Supervised Imitation Learning For Long-term Microscopic Traffic Simulation (CVPR 2024 · similarity 0.88)
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.88)
- SceneDiffuser++: City-Scale Traffic Simulation via a Generative World Model (CVPR 2025 · similarity 0.88)
- Transfer Your Perspective: Controllable 3D Generation from Any Viewpoint in a Driving Scene (CVPR 2025 · similarity 0.87)
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.
