A GAN (Generative Adversarial Network) model for generating more optimal paths that closely resemble those produced by the A* algorithm when conditioned on a 2D obstacle map and an initial RRT (Rapidly-exploring Random Tree) path.
To collect 2D RRT and A* plans generated on random environments with random start and end goal positions, please refer to the script located at PathPlanning/collect_data.py. For details on the training pipeline, see the Modeling/Plan2Plan_Transformer.ipynb script.