Efficacy of Early and Late Spatio-Temporal Fusion in LSTM Models for Vehicle Trajectory Prediction
Bibliographic record
Abstract
Vehicle trajectory prediction is crucial for optimizing transportation systems, reducing traffic congestion, minimizing travel time, and enhancing overall safety. Accurately forecasting trajectories in complex traffic remains challenging due to both temporal vehicle dynamics and spatial interactions among neighboring vehicles. In this paper, we present a head-to-head comparison of two long short-term memory (LSTM) architectures that fuse spatio-temporal data at different stages: an early-fusion model that aggregates neighbor features at every time step, and a late-fusion model that encodes target and neighbor sequences separately and merges them only at the final prediction layer. A vanilla LSTM baseline that uses only the target vehicle's own history (no neighbor information) is also included to isolate the fusion benefit. All models are trained and evaluated on the NGSIM US-101 highway dataset using mean squared error (MSE), and mean absolute error (MAE) metrics. The results show that both fusion strategies yield modest error reductions compared to the baseline (vanilla LSTM MSE <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\boldsymbol{=} \mathbf{1. 6 5} \boldsymbol{\rightarrow}$</tex> early-fusion MSE <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\boldsymbol{=} \mathbf{0. 9 8 7 4}$</tex>, late-fusion MSE <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$=0.9910$</tex>), with only a slight advantage for early fusion. We provide qualitative analyses of lane-change versus lane-keep scenarios, discuss the limitations of our one-step prediction and simple neighbor pooling, and outline directions for future work, including attention mechanisms and graph-based models. The data and implementation are available at https://github.com/sukhjitsehra/FVTP.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.001 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".