2
nuReasoning contains 20K 20-second long-tail driving data clips from real-world scenarios.
nuReasoning is a large-scale real-world long-tail driving dataset containing 20K 20-second clips across diverse scenario types. The dataset provides high-quality reasoning annotations spanning spatial reasoning, driving decisions, and counterfactual reasoning. Compared with prior datasets, nuReasoning offers substantially larger-scale long-tail driving data and richer reasoning annotations, enabling models trained on it to achieve significantly improved reasoning and planning performance.
Models trained on nuReasoning significantly improve reasoning performance across all four core capabilities in driving tasks compared to both base models and other general-purpose models.
The nuVLA model outperforms state-of-the-art AV planning methods, while training with all reasoning types (spatial, driving, and counterfactual) yields the best overall planning performance.
@misc{nureasoning2026,
title = {nuReasoning: A Reasoning-Centric Dataset and Benchmark for Long-Tail Autonomous Driving},
author = {Huang, Zhiyu and Liu, Johnson and Song, Rui and Zhou, Zewei and Yang, Ruining and Zhang, Yun and Cai, Tianhui and Zhang, Hanyin and Gao, Mingxuan and Xu, Valeria and Chen, Jiali and Shen, Yishan and Guo, Yiluan and Qi, Xuewei and Ma, Jiaqi},
year = {2026},
note = {Placeholder citation.}
}