A non-local multi-head spatiotemporal attention LSTM for vehicle trajectory prediction
Sidra Rashid, Muazzam A Khan, Usman Akram, Raheel Mumtaz, It Ee Lee, Toqeer Ali Syed · 2026
Read A non-local multi-head spatiotemporal attention LSTM for vehicle trajectory prediction by Sidra Rashid, Muazzam A Khan, Usman Akram, Raheel Mumtaz, It Ee Lee, Toqeer Ali Syed (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/plos-pone-0357735.txt.
Description
Sidra Rashid leads a team of six teaching self driving systems to guess where cars will go next, in PLOS ONE in 2026. A vehicle's future path depends on its neighbors, but not equally: one close truck matters more than five distant sedans. Earlier models either watched a small neighborhood or treated every neighbor alike, and their errors grew over longer horizons. The team's model pairs a memory network with a multi head attention system that ranges across the scene.
Attention scores let the model weight the vehicles that influence the target, and position aware features capture long range patterns. On a public highway dataset, the model beat every established rival. At the five second horizon its prediction error fell by more than half compared with the next best method. For a car planning a lane change, that margin is the difference between a calm decision and a late swerve.
Catalog record
| CREATOR | Sidra Rashid, Muazzam A Khan, Usman Akram, Raheel Mumtaz, It Ee Lee, Toqeer Ali Syed |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47995 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |