Recall to Predict: Grounding Motion Forecasting in Interpretable Motion Bank
Abhishek Vivekanandan, Ahmed Abouelazm, J. Marius Zöllner · 2026
Read Recall to Predict: Grounding Motion Forecasting in Interpretable Motion Bank by Abhishek Vivekanandan, Ahmed Abouelazm, J. Marius Zöllner (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-01393.txt.
Description
Abhishek Vivekanandan, Ahmed Abouelazm, and J Marius Zöllner ground self-driving motion forecasts in a bank of real, readable example paths. Standard systems either hide their guesses in opaque learned queries, which can collapse to sameness, or sample paths naively, which limits variety.
Their framework builds a motion bank, a structured space of physically possible paths learned by contrast. At run time, a retrieval layer picks fitting example paths from the bank using adapted queries and a sampling trick that keeps training smooth. A decoder then refines the picked paths, trained with a winner-takes-all mixture model and extra terms that reward variety and close endpoints. Because each forecast starts from a named, sensible path type, the output stays open to inspection. On the Argoverse 2 and Waymo motion data sets, scores stay competitive with the best opaque systems.
Catalog record
| CREATOR | Abhishek Vivekanandan, Ahmed Abouelazm, J. Marius Zöllner |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.01393