LAIA Dataset: Synthetic Human Attention Data for Autonomous Driving
LAIA (Labelled Attention for Intelligent Automobiles) is a newly developed synthetic dataset aimed at enhancing the interpretability of end-to-end autonomous driving systems. This dataset, gathered from the CARLA simulator in controlled environments, features more than 15 hours of driving data contributed by 44 participants across meticulously designed scenarios. Each recorded sequence offers RGB images under six different weather conditions, along with semantic and instance segmentation, depth information, optical flow, and CAN bus data. By supplying human attention data, the dataset seeks to tackle explainability issues, facilitating a deeper comprehension of AI decision-making processes in autonomous vehicles.
Key facts
- LAIA stands for Labelled Attention for Intelligent Automobiles
- Dataset collected using CARLA simulator in closed-loop environments
- Over 15 hours of driving data from 44 participants
- Includes RGB images under six weather conditions
- Includes semantic and instance segmentation, depth, optical flow, and CAN bus data
- Designed to enrich end-to-end driving research with human attention data
- Addresses challenges in interpretability and explainability of end-to-end driving paradigms
- Published on arXiv with ID 2607.25570
Entities
Institutions
- arXiv