This repository is about my learning journey on the whole pipeline of a machine learning project. It is done based on this research paper. For more information, please refer to that paper and github repository.
The model used in this project is GNN-RNN for single vehicle trajectory prediction.
Refer to this github repository.
python training.py
Legend:
grey solid line: history trajectory of surrounding vehicle (3s)
red solid line: history trajectory of ego vehicle (3s)
blue dotted line: ground truth of future trajectory of ego vehicle (5s)
green solid line: predicted future trajectory of ego vehicle (5s)
Remember, garbage in, garbage out. At first, I only prepare the dataset with 700+ examples, which is definitely not enough. The model is overfit after training and the error never goes down steadily. I am not aware of this and spend weeks to debug until I realised there is no bugs in my code.