RACER (Rationale-Aware Captioning of Edge-Case Driving Scenarios) is a reasoning caption dataset designed for training vision-language-action (VLA) models in autonomous driving.
This repository provides approximately 1,000 samples, as a small subset of the RACER dataset. Each sample consists of a temporal sequence of front camera images, the ego vehicle’s future trajectory, and a corresponding reasoning caption.
For details, please refer to our techblog RACER:… See the full description on the dataset page:
https://huggingface.co/datasets/turing-motors/RACER-Mini.
