Title: Navigating the Future of Self-Driving: An Introduction to the KITScenes Multimodal Dataset

While current autonomous driving datasets have driven significant advancements, they often lack sufficient sensor precision, comprehensive mapping, or geographic variety. To address these limitations, we introduce KITScenes Multimodal, a European dataset centered on high-fidelity sensors and maps.

Our sensor suite is fully synchronized and includes high-resolution global-shutter cameras, 4D imaging radar, redundant GNSS/INS localization, and long-range LiDAR capable of detecting objects beyond 400 meters. To the best of our knowledge, this dataset features the most complete HD maps available in any sensor-based collection. Their accuracy has been confirmed through autonomous driving trials using open-source software.

In a first for public datasets, we map all traffic-relevant elements—including traffic lights—in 3D with reprojection accuracy and full topological connectivity. By recording data in cities characterized by irregular street layouts and mixed traffic modes, KITScenes expands the geographic diversity available to researchers.

Additionally, we present four new benchmarks designed to advance spatial learning for embodied AI: * Online HD map construction * Long-range depth estimation * Novel view synthesis * End-to-end driving

For more information, visit the project page: https://kitscenes.com/

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC