Bench2Drive
Bench2Drive is the first benchmark for evaluating E2E-AD systems’ multiple abilities in a closed-loop manner. Bench2Drive consists of 2 million fully annotated frames as official training data, collected from 13638 short clips by World Model RL expert Think2Drive, uniformly distributed under 44 interactive scenarios (cut-in, overtaking, detour, etc), 23 weathers (sunny, foggy, rainy, etc), and 12 towns (urban, village, university, etc) in CARLA Leaderboard v2. Its evaluation protocol requires E2E-AD models to pass those 44 interactive scenarios under different locations which sums up to 220 routes and thus provide a comprehensive and disentangled assessment about their driving capability under different situations. We implement state-of-the-art E2E-AD models and evaluate them in Bench2Drive to provide more insights regarding current status of E2E-AD.
Key Features:
- Comprehensive scenario coverage: 100k clips, 44 scenarios, 23 weathers, 12 towns.
- Environment Evaluation from World Model RL expert - Think2Drive: intermediate features, reinforcement learning reward, value and action.
- Multi sensors (camera, LiDAR, Radar, etc) and rich annotation information: 2D/3D boxes, depth, semantic/instance segmentation, HD-Map
- Granular skill assessment and closed-Loop evaluation protocol
What does Think2Drive + Bench2Drive provide ?
Diverse Sensor Anno
Challenging Scenes Collected By World Model RL Expert Think2Drive
Sensor Suites
- 1x LiDAR: 64 Channels, 85m range, 600,000 points per second
- 6x Camera: 70° & 110° FoV, 900×1600 Resolution
- 5x Radar: 100m range, 30° horizontal and vertical FoV
- 1x IMU&GNSS
Data Distribution
Scenario Distribution of Bench2Drive Dataset.
Town and Weather Distribution of Bench2Drive Dataset.
Benchmark
