Argoverse 1 Motion Forecasting is a curated collection of 324,557 scenarios, each 5 seconds long, for training and validation. Each scenario contains the 2D, birds-eye-view centroid of each tracked object sampled at 10 Hz.
To create this collection, we sifted through more than 1000 hours of driving data from our fleet of self-driving test vehicles to find the most challenging segments — including segments that show vehicles at intersections, vehicles taking left or right turns, and vehicles changing lanes.
What makes this dataset stand out?
This dataset is far larger than what can currently be mined from publicly available self-driving datasets, and our HD maps make it easier to predict the motion of objects.
Segment duration:
5 seconds
Total number of segments:
323,557
A still frame from one of the motion forecasting sequences, showing the trajectories for the agent of interest (red), self-driving vehicle (green), and all other objects of interest in the scene (light blue).