June 2024
tl;dr: Query centic prediction that marries agent centric and scene centric predictions.
Overall impression
Winning solution in Argoverse and Waymo datasets.
Key ideas
- Local coordinate system for each agent that leverages invariance.
- Long horizon prediction in 6-8s is achieved by AR decoding of 1s each, then followed by a trajectory refiner. –> This means the target oriented approach scuh as TNT might have been too hard. TNT seems to have been proposed to maximize FDE directly.
Technical details
- Summary of technical details, such as important training details, or bugs of previous benchmarks.
Notes