Code for running "Simultaneous Mapping and Target-driven Navigation" in Habitat simulator.

Installation

1. Install habitat-lab and habitat-sim from the following repository.
   https://github.com/facebookresearch/habitat-lab

2. The following libraries should also be necessary.
   torch >= 1.5
   torchvision == 0.6
   networkx
   visualpriors
   

3. Put my_objectnav_mp3d.yaml, my_objectnav_mp3d_train.yaml, my_objectnav_mp3d_test.yaml into the data folder.

Habitat Code Example

1. examples/shortest_path_follower_example.pywill draw a top down map visualizing the agent following a shortest path to the goal. It is also an example of running shortest path on the pointNav task.

2. temp_geodesic_follower.py is an example for running shortest path on the objectNav task.

Training and Testing

All the hyper-parameters are defined in parameters_habitat.py.
MapNet module and navigation module are trained and tested separately.

1. To train and test the MapNet module,
   python train_MapNet_habitat.py
python test_MapNet_habitat.py

2. To train and test the navigator,
   python train_IL_habitat.py
python test_IL_habitat.py