New KITScenes LongTail dataset tackles rare driving scenarios with multilingual reasoning traces to improve autonomous vehicle generalization

Dataset focuses on long-tail driving events with multi-view video, trajectories, high-level instructions, and detailed reasoning traces for end-to-end driving

Includes multilingual reasoning traces in English, Spanish, and Chinese from domain experts with diverse cultural backgrounds

Designed to evaluate Vision Language Models (VLMs) and Vision Language Action models (VLAs) on instruction following and semantic coherence beyond safety metrics

Supports in-context learning and few-shot generalization to improve real-world autonomous driving in rare and uncommon scenarios

Dataset publicly available on Hugging Face at https://hf.co/datasets/kit-mrt/kitscenes-longtail