Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
End-to-End Autonomous Driving Through V2X Cooperation
Authors: Haibao Yu, Wenxian Yang, Jiaru Zhong, Zhenwei Yang, Siqi Fan, Ping Luo, Zaiqing Nie
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate the effectiveness of Uni V2X in significantly enhancing planning performance, as well as all intermediate output performance. |
| Researcher Affiliation | Academia | 1The University of Hong Kong 2AIR, Tsinghua University 3Beijing Institute of Technology 4University of Science and Technology Beijing |
| Pseudocode | No | The paper describes the proposed method, Uni V2X, using textual descriptions and figures (e.g., Figure 2 for the pipeline), but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code https://github.com/AIR-THU/Uni V2X |
| Open Datasets | Yes | We implement Uni V2X, as well as reproducing several benchmark methods, on the challenging DAIR-V2X, the real-world cooperative driving dataset. ... as well as instantiating the Uni V2X on DAIRV2X (Yu et al. 2022). ... We also conduct Uni V2X on more V2X datasets such as V2X-Sim (Li et al. 2022a), and present the experiment results in the Appendix. |
| Dataset Splits | No | DAIR-V2X Dataset comprises approximately 100 scenes captured at 28 complex traffic intersections, recorded using both infrastructure and vehicle sensors. Each scene has a duration ranging from 10 to 25 seconds, capturing data at a rate of 10 Hz, and is equipped with a high-definition (HD) map. This dataset provides a diverse range of driving behaviors, including actions such as moving forward, turning left, and turning right. To align with nu Scenes (Caesar et al. 2020), we categorize object classes into four categories (car, bicycle, pedestrian, traffic cone). However, specific training, validation, or test dataset splits (e.g., percentages or sample counts) are not provided in the main text. |
| Hardware Specification | Yes | The experiments are conducted utilizing 8 NVIDIA A100 GPUs. |
| Software Dependencies | No | The paper mentions architectural components like BEVFormer, DETR, and Panoptic Seg Former as bases for its modules but does not specify software dependencies with version numbers (e.g., Python, PyTorch, or CUDA versions). |
| Experiment Setup | Yes | We establish the interest range of the ego vehicle as [-50, 50, -50, 50] meters. The ego-vehicle BEV range shares the same area spanning [-50, 50, -50, 50] meters, with each grid measuring 0.25m by 0.25m. The infrastructure BEV range is set as [0, 100, -50, 50] meters... In Baseline Settings... we remove the ego-vehicle velocity embedding in all baseline settings for a fair comparison. |