API
Download
Preview May 2026
KITScenes Multimodal Early Release
A preview of the Multimodal dataset is available on HuggingFace. This early release is intended for pipeline exploration; it is not recommended for final benchmark reporting.
Autonomous Driving Datasets
Advancing end-to-end driving from research to safe autonomy — from high-fidelity multimodal sensing and production-grade HD maps to rare long-tail events that require complex semantic reasoning.
Institute of Measurement and Control Systems (MRT)
Karlsruhe Institute of Technology (KIT)
KITScenes is a collection of autonomous driving datasets published by the Institute of Measurement and Control Systems (MRT) at the Karlsruhe Institute of Technology (KIT) . Our dataset tracks address the full spectrum of challenges on the road to L4 autonomy — from holistic multimodal spatial reasoning and HD map perception to safe behavior in rare and unforeseeable long-tail situations.
1,000 rare driving scenarios — construction zones, adverse weather, nighttime, and overtaking — each captured as a nine-second 360° video clip with trajectories and high-level instructions, and expert reasoning traces in English, Spanish, and Chinese. The dataset facilitates in-context learning, few-shot generalization, and research on cross-lingual driving reasoning.
A high-fidelity sensor suite paired with production-grade Lanelet2 HD maps covering all elements required for real-world L4 autonomous driving, validated for closed-loop use with the open-source Autoware stack. Together they ground multimodal spatial reasoning in the physical world — enabling research on long-range perception, novel view synthesis, and physical AI that drives safely in real environments.
Latest releases, papers, and announcements.
A preview of the Multimodal dataset is available on HuggingFace. This early release is intended for pipeline exploration; it is not recommended for final benchmark reporting.
Test split and 3 training samples are now available on HuggingFace for few-shot evaluation. Full training and validation splits will follow.
Stay tuned for more updates to come!
