End-to-End
Learn trajectories or controls directly from sensor input, with planning-oriented representations, closed-loop performance, and real-time execution.
CURATED · REPRODUCIBLE · OPEN
Tracking the next
driving intelligence
A curated map of end-to-end driving, world models, and vision-language-action research, connected to public protocols, code, and reproducible results.
PLANNING ✦ GENERATION ✦ REASONING ✦ CLOSED-LOOP ✦ SAFETY ✦ LANGUAGE-GROUNDED ACTION ✦
01
RESEARCH LANDSCAPE
Four research tracks.
One evidence-based hub.
World Model
Learn spatial-temporal driving dynamics for video generation, future prediction, data scaling, simulation, and policy learning.
Vision · Language · Action
Unify visual understanding, language reasoning, and driving actions for long-tail scenarios, interaction, and interpretable decisions.
Recursive Self-Improvement
Study driving policies and agents that improve through repeated feedback, safe learning, real-world experience, and evolving scenarios.
02
CURATED PAPERS
Read beyond the abstract.
Browse, save, and open a structured reading note for each paper. Notes link to their sources and separate indexed evidence from questions to verify in the full text.
No papers match these filters.
Try a broader search or reset the filters.
03
DATASETS & BENCHMARKS
Public resources with traceable protocols
PUBLIC DATA SOURCES
Curated autonomous-driving datasets
Browse three broad groups: perception and multimodal understanding, motion, planning and cooperative driving, and simulation and synthetic data. Each entry shows its method or dataset name, associated paper title, scale, access terms, and source links.
EVALUATION PROTOCOLS
Public test benchmarks
Explore closed-loop planning, open-loop and non-reactive planning, VLA reasoning, and reliability benchmarks. Each leaderboard preserves its original protocol, data split, and primary metric; scores from incompatible protocols stay separate.
04
COMMUNITY LEADERBOARD
An auditable public leaderboard
Every result must link to code, configuration, and primary evaluation evidence.
| RANK | METHOD | TEAM | SPLIT | METRIC | STATUS | EVIDENCE |
|---|
SUBMIT YOUR RESULT
Make results reproducible and comparisons trustworthy.
- Copy
benchmark/submissions/example.json - Add the protocol, score, and evidence
- Run
npm run validate - Open a pull request
05
EDITORIAL POLICY
Traceable sources
Prefer arXiv, conference pages, official repositories, and dataset websites.
Protocol integrity
Separate open-loop, closed-loop, reactive, and non-reactive evaluation.
Auditable results
Every leaderboard entry links its configuration, date, and evaluation evidence.