Learning-based motion prediction for autonomous vehicles using sequence models (LSTM) and a feed-forward baseline (MLP). The project provides a full pipeline: data loading → preprocessing → training → evaluation → visualization, with reproducible configs.
In the experiments, the LSTM achieved ~30.5% lower trajectory prediction error than the baseline MLP.
Key Features
-
Trajectory prediction from past motion histories (positions/velocities)
-
Models: LSTM and MLP baseline
-
Visualizations: predicted vs. ground-truth trajectories, error heatmaps