By James Hays, Principal Scientist, Argo AI
Earlier this year, we released Argoverse, a curated collection of high-definition maps and sensor data from a fleet of Argo AI self-driving test vehicles. Today, we’re launching the first two Argoverse competitions using our motion forecasting and 3D tracking datasets, and inviting academic researchers and students to participate.
To support the competitions, we’re opening two evaluation servers to enable participants to benchmark their 3D tracking and motion forecasting methods, and see how their methods stack up against competitors’ submissions. These servers will be open for submissions through December 1, 2019, and the competition will close with a session at the NeurIPS 2019 Workshop on Machine Learning for Autonomous Driving on December 14, 2019.
The first place winning entry for each task will receive a $1,500 cash prize and be invited to intern with Argo in Summer 2020*. Second place winners will receive a cash prize of $500. In addition, we will award $250 as an honorable mention to teams with the most creative entries.
We released the Argoverse data collection in June 2019 to aid the academic community in making advancements in key perception and forecasting tasks for self-driving technology, and to provide resources to explore the impact of high-definition maps on these tasks (for more on the Argoverse data collection, see our blog). But our commitment to sharing high-quality, curated data and tools with the academic community didn’t end there. Coinciding with these competitions and in response to feedback, we’ve made a few improvements to our datasets, including improvement of stereo calibration and 3D track annotations, and removal of noisy motion forecasting sequences. We refer to these updates as Argoverse 1.1.
The 3D tracking and motion forecasting competitions are the first in a series of Argoverse competitions. As we continue to make additions to our datasets and offer more resources to the academic community, we will launch additional opportunities to compete through tasks that are critical to advancing the ways self-driving vehicles understand the world around them. To stay up to date on those additions and more, sign up for the Argoverse mailing list.
So, what are you waiting for? Head on over to www.argoverse.org/tasks to get started.