Large-scale open source dataset for autonomous driving.
nuScenes is an initiative intended to support research to further advance the mobility industry.
With this goal in mind, the dataset includes 1000 scenes collected in Boston and Singapore and is the largest multi-sensor dataset for autonomous vehicles.
It features:
For the dataset, Aptiv managed the collection of data, carefully choosing to capture challenging scenarios and a diversity of locations, times and weather conditions.
Collected in Boston's Seaport and Singapore's One North, Queenstown and Holland Village districts, each of the 1000 scenes in the dataset were manually selected.
Two identical cars with identical sensor layouts were used to drive in Boston and Singapore. Refer to the image below for the placement of the sensors:
Careful calibration of the extrinsics and intrinsics of every sensor is critical to achieving a high quality dataset. For nuScenes, extrinsic coordinates are expressed relative to the ego frame (ie: the midpoint of the rear vehicle axle).
We use a laser liner to accurately measure the relative location of the LiDAR to the ego frame.
We place a cube-shaped calibration target in front of the camera and LiDAR sensors. The calibration target consists of three orthogonal planes with known patterns. After detecting the patterns we compute the transformation matrix from camera to LiDAR by aligning the planes of the calibration target. Given the LiDAR to ego frame transformation computed above, we can then compute the camera to ego frame transformation and the resulting extrinsic parameters.
We mount the radar in a horizontal position. Then we collect radar measurements by driving in an urban environment. After filtering radar returns for moving objects, we calibrate the yaw angle using a brute force approach to minimize the compensated range rates for static objects.
We use a calibration target board with a known set of patterns to infer the intrinsic and distortion parameters of the camera.
In order to achieve cross-modality data alignment between the LiDAR and the cameras, the exposure on each camera was triggered when the top LiDAR sweeps across the center of the camera’s FOV. This method was selected as it generally yields good data alignment.
Note that the cameras run at 12Hz while the LiDAR runs at 20Hz. The 12 camera exposures are spread as evenly as possible across the 20 LiDAR scans, so not all LiDAR scans have a corresponding camera frame.
Reducing the frame rate of the cameras to 12Hz helps to reduce the compute, bandwidth and storage requirement of the perception system.
As Aptiv's partner in developing nuScenes, Scale contributed two things: First, is the data annotation, including deciding on the taxonomy and developing instructions for labelers and managing QA.
Second is Scale's web-based visualizer for LiDAR and camera data for exploring the dataset. Scale's visualizer allows point cloud data to be easily embedded into any webpage and shared.
Ready to get started with nuScenes? This tutorial will give you an overview of the dataset without the need to download it. Please note that this page is a rendered version of a Jupyter Notebook.