A2RL Vmax at a glance

Competition coverage

Data from time-trial, attack-and-defend, and the final four-vehicle race at Yas Marina F1 Circuit.

297 scenes and 466 tracks

Multi-vehicle interaction data curated for detection, tracking, and future prediction tasks.

305,531 LiDAR point clouds

Includes 28,791 professionally annotated frames and 38,545 vehicle bounding boxes.

386,006 RADAR point clouds

Complementary long-range sensing for extreme-speed perception research.

Eight contributing teams

Competition data shared across participating teams to build a common research benchmark.

nuScenes-compatible format

Designed for straightforward training and evaluation with existing 3D detection and tracking tooling.

A2RL Vmax: The A2RL autonomous racing dataset for long-range, high-speed perception and multi-vehicle interaction

Marvin Klemp1,10, Dominic Ebner2, Cornelius Schröder2, Davide Malvezzi4, László Turányi5, Riccardo Donati6, Ilia Schminik7, Xia Ning8, Yanxin Zhou9, Matthew Flagg11, Christoph Stiller1, Markus Lienkamp2, Marko Bertogna4, Gergely Bári5, Andreas Birk7, Ren Jin8, Chen Lv9, Johannes Betz3
1Karlsruhe Institute of Technology 2Institute of Automotive Technology, Technical University of Munich 3Professorship of Autonomous Vehicle Systems, Technical University of Munich 4University of Modena and Reggio Emilia 5Széchenyi István University 6Polytechnic Institute of Milan 7Constructor University 8Beijing Institute of Technology 9Nanyang Technological University 10AIPEX 11Code 19 Racing
IEEE International Conference on Intelligent Robots and Systems (IROS) 2026
Paper Dataset Read Online BibTeX
A2RL Vmax dataset overview showing LiDAR, RADAR, IMU and wheel-speed signals, and the Yas Marina F1 Circuit

A2RL Vmax introduces long-range, high-speed perception data collected during the Abu Dhabi Autonomous Racing League, including multi-vehicle race interactions and professionally annotated LiDAR point clouds.

Abstract

A2RL Vmax is an open-source perception dataset built for high-speed autonomous driving and multi-vehicle racing interaction. The data was captured during the 2024 Abu Dhabi Autonomous Racing League at the Yas Marina F1 Circuit and covers both single-vehicle sessions at varying speeds and competitive multi-vehicle race scenarios.

The dataset contains nearly 30,000 professionally annotated LiDAR point clouds together with RADAR point clouds, making it the first large-scale autonomous racing dataset of this kind to support deep-learning-based perception research with dense expert labels.

To make the dataset practical for future work, the authors provide the data in a developer-friendly format and benchmark off-the-shelf 3D detection and tracking methods, showing that while standard baselines perform reasonably well at short range, specialized methods are still needed for long-range, high-speed racing conditions.

Data Highlights

Short visual glimpses into the sensing and motion data captured in A2RL Vmax.

Localization

Localization and state estimation highlight from the A2RL dataset

Vehicle Data

Vehicle data highlight with onboard camera, LiDAR and motion signals

LiDAR Perception

LiDAR highlight showing long-range sensing and ego-vehicle view

Figures

Three core visuals from the paper summarizing the race platform, sensing stack, and benchmark setting.
Two A2RL autonomous racecars competing on track

Competitive multi-vehicle racing at A2RL, highlighting the real race setting behind the dataset.

A2RL autonomous racecar close-up at Yas Marina Circuit

The A2RL autonomous racecar platform used to capture high-speed perception data.

Satellite view of the Yas Marina F1 Circuit

A satellite view of Yas Marina Circuit, the venue for the 2024 Abu Dhabi Autonomous Racing League.

LiDAR point cloud example from the A2RL dataset

Example LiDAR point cloud showing the long-range sensing conditions present in the dataset.

RADAR point cloud example from the A2RL dataset

Example RADAR point cloud illustrating complementary sparse sensing for racing perception.

Key Contributions

High-speed perception benchmark

The dataset focuses on long-range perception and multi-vehicle interaction under genuine racing constraints rather than everyday road traffic.

Professionally annotated LiDAR

A2RL Vmax provides large-scale expert labels for 3D perception research in autonomous racing.

Practical evaluation tooling

The paper includes baseline detection and tracking results and packages the data in a nuScenes-compatible format.

Dataset Scope

297 scenes
466 opponent tracks
305,531 LiDAR point clouds
28,791 annotated LiDAR frames
38,545 vehicle bounding boxes
386,006 RADAR point clouds

Paper

BibTeX

@inproceedings{klemp2026a2rlvmax,
  title={A2RL Vmax: The A2RL autonomous racing dataset for long-range, high-speed perception and multi-vehicle interaction},
  author={Klemp, Marvin and Ebner, Dominic and Schr{\"o}der, Cornelius and Malvezzi, Davide and Tur{\'a}nyi, L{\'a}szl{\'o} and Donati, Riccardo and Schminik, Ilia and Ning, Xia and Zhou, Yanxin and Flagg, Matthew and Stiller, Christoph and Lienkamp, Markus and Bertogna, Marko and B{\'a}ri, Gergely and Birk, Andreas and Jin, Ren and Lv, Chen and Betz, Johannes},
  booktitle={IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  year={2026},
  url={https://tum-avs.github.io/A2RL_Dataset_website/}
}