This is the implementation of RRT*former
A suitable conda environment named RRTformer can be created and activated with:
conda env create -f environment.yaml
conda activate RRTformerIf you want to record the training process, wandb is also needed.
Download rrt_net_data.zip and move the zip file into the root folder of this repo. Run
cd RRTformer/
unzip rrt_net_data.zip- For 2D, run
cd 2d
python demo_planning.py --env_file='../data/random_2d/test/envs_fsg.json'env_file is the json file of your environment. Visualization will be shown in GUI.
- For 3D, run
cd 3d
python demo_planning_3d.py --env_file='../data/random_3d/test/envs_fsg.json'Instructions for generate your own data for train or test.
cd 2d
python gene_data_json.py --yaml_path='../data/random_2d/random_2d.yml' --json_path='../data/random_2d/test/envs.json' --mode='test'
python gene_data_astar.py --envs_file='../data/random_2d/train/envs.json' # For trainFor training data, mode=train, for testing data, mode=test. gene_data_json is to generate the random environment. gene_data_star is to generate A* path in an environment for training.
cd 3d
python gene_data_json_3d.py --yaml_path='../data/random_3d/random_3d.yml' --json_path='../data/random_3d/test/envs.json' --mode='test'
python gene_data_astar_3d.py --envs_file='../data/random_3d/train/envs.json' # For trainInstructions for training your own models.
Run
cd 2d
python train.py --dataset_path='../data/astar_2d.npy' --batch_size=256 --epochs=2000 --lr=0.0001 --checkpoint_dir='../checkpoint/2d'For better results, the relevant parameters can be changed.
Run
cd 3d
python train_3d.py --dataset_path='../data/astar_3d.npy' --batch_size=128 --epochs=2000 --lr=0.0001 --checkpoint_dir='../checkpoint/3d'Run
cd 2d
python evaluate.py --env_file='../data/random_2d/test/envs_fsg.json'Run
cd 3d
python evaluate_3d.py --env_file='../data/random_3d/test/envs_fsg.json'Put files in evaluation to nirrt_star/results/evaluation + 2d or 3d in tedhuang96/nirrt_star and follow nirrt_star's instructions.