FOLDER_NAME = 'viper' model_path = f'model/{FOLDER_NAME}' train_path = f'train/{FOLDER_NAME}' gifs_path = f'gifs/{FOLDER_NAME}' SUMMARY_WINDOW = 2 LOAD_MODEL = False # load trained model and resume training SAVE_IMG_GAP = 1000 N_AGENTS = 4 EXPLORATION = True # True: unknown map, False: known map CELL_SIZE = 0.4 # meter per pixel NODE_RESOLUTION = 4.0 # meter of node interval DOWNSAMPLE_SIZE = NODE_RESOLUTION // CELL_SIZE SENSOR_RANGE = 20 # meter, 7.9812 for Gregorin's maps UTILITY_RANGE = 0.8 * SENSOR_RANGE EVADER_SPEED = SENSOR_RANGE MIN_UTILITY = 0 FRONTIER_CELL_SIZE = 4 * CELL_SIZE LOCAL_MAP_SIZE = 40 # meter EXTENDED_LOCAL_MAP_SIZE = 8 * SENSOR_RANGE * 1.05 MAX_EPISODE_STEP = 128 REPLAY_SIZE = 10000 MINIMUM_BUFFER_SIZE = 5000 BATCH_SIZE = 256 LR = 2e-5 GAMMA = 1 NODE_INPUT_DIM = 8 EMBEDDING_DIM = 128 LOCAL_K_SIZE = 25 # the number of neighboring nodes LOCAL_NODE_PADDING_SIZE = 360 # the number of nodes will be padded to this value USE_GPU = False # do you want to collect training data using GPUs USE_GPU_GLOBAL = True # do you want to train the network using GPUs NUM_GPU = 1 NUM_META_AGENT = 24 # number of parallel environments USE_WANDB = False