# Thinklab-SJTU/Bench2Drive

[NeurIPS 2024 Datasets and Benchmarks Track] Closed-Loop E2E-AD Benchmark Enhanced by World Model RL Expert

Repository: https://github.com/Thinklab-SJTU/Bench2Drive
Canonical: https://ross.abutalabs.com/products/bench2drive
Language: Python
License: NOASSERTION
License Family: other
Last push: 2026-08-11T04:01:59+00:00

## Health v2 (maintenance only)
Score: 68/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 97, release rhythm 35, longevity 61
- inputs: {"age_days": 862, "days_push": 22, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1926, forks 142 (observed 2026-08-28T04:05:55.564304+00:00)

## What it is
Bench2Drive is a closed-loop benchmark and dataset for end-to-end autonomous driving, built on CARLA with an RL-based expert driver (Think2Drive). It provides multi-sensor driving data across 44 scenarios and evaluation tooling for E2E-AD models.

## Use cases
- benchmark end-to-end autonomous driving models in closed loop
- download driving datasets for training AD models
- evaluate driving policies on CARLA scenarios
- train imitation learning agents from expert demonstrations
- run ablation studies on a small validation subset
- test robustness of driving models to sensor failures

## When to choose
- you need a standardized closed-loop benchmark for end-to-end autonomous driving research
- you want uniformly distributed scenario training data with 3D occupancy labels
- you need an RL expert for demonstration data in CARLA

## When to avoid
- you need real-world driving data rather than CARLA simulation
- you work outside Python 3.7/3.8 CARLA constraints without the protocol bridge
- you lack GPU resources for large-scale simulation and training

## Facets
- artifact type: dataset
- maturity: active
- function: benchmarking, machine-learning, simulation, data-generation
- domain: autonomous-vehicles, machine-learning, simulation, computer-vision
- platform: python
- tags: autonomous-driving, carla, end-to-end-driving, closed-loop-benchmark, world-model, reinforcement-learning-expert, neurips-2024, 3d-occupancy, linux, docker

## Member repositories
- Thinklab-SJTU/Bench2Drive (main) score 68

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:55.564304+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:09:12.725358+00:00, confidence not recorded.
  - readme: https://github.com/Thinklab-SJTU/Bench2Drive (fetched 2026-08-28T04:05:55.564304+00:00, sha 293457fce016)
- Data as of 2026-08-30T08:39:29.467469+00:00.