Autonomous driving research depends heavily on data. The more accurately researchers can capture what a vehicle sees, where objects are located and how they are moving, the better they can develop and evaluate perception systems. Aeva's release of AevaScenes takes an interesting step in this direction by making a large dataset combining FMCW 4D LiDAR and camera data freely available for academic and non-commercial research. 

The dataset brings together synchronised measurements from six FMCW LiDAR sensors and six high-resolution RGB cameras. Unlike conventional LiDAR datasets that primarily provide information about distance, FMCW LiDAR can capture both position and velocity. This additional information is particularly valuable for research into object detection, tracking, motion forecasting and scene understanding. AevaScenes also includes semantic segmentation and lane-line annotations, giving researchers several layers of information with which to train and evaluate perception algorithms. 

The scale of the dataset is another notable aspect. It contains 100 driving sequences and around 10,000 synchronised frames, covering both urban and highway environments as well as daytime and night-time conditions. The annotations extend to distances of up to 400 metres, which could be particularly useful for studying how autonomous vehicles perceive objects and traffic situations further ahead. 

Open datasets such as this can help make autonomous driving research more reproducible. Researchers working on sensor fusion, for example, can work with the same underlying data when comparing different algorithms instead of relying exclusively on proprietary datasets. That can make it easier to benchmark approaches and identify where particular perception methods perform well or struggle.

At the same time, the dataset has clear limitations: the sequences were collected around the San Francisco Bay Area, in clear weather and on dry roads. It therefore cannot represent the full diversity of conditions encountered by vehicles in Europe, including snow, heavy rain or different road infrastructure.

Making high-quality multimodal sensor data openly available is nevertheless an important contribution to the development of automated driving. Better shared datasets can help researchers test new ideas more systematically and bring greater transparency to the rapidly evolving field of vehicle perception.

You can read the full article here. AEVA. Aeva Introduces AevaScenes, the First Open-Access FMCW 4D LiDAR and Camera Dataset for Autonomous Vehicle Research [online]. September 30, 2025 [cited 2026-08-16]. Available from: https://www.aeva.com/press/aeva-introduces-aevascenes-the-first-open-access-fmcw-4d-lidar-and-camera-dataset-for-autonomous-vehicle-research/

Picture: chat gpt