LearningToNavigate

Preparing conda env

Assuming you have conda installed, let's prepare a conda env:

# We require python>=3.7 and cmake>=3.10
conda create -n habitat python=3.7 cmake=3.14.0
conda activate habitat

conda install habitat-sim

• To install habitat-sim with bullet physics

  conda install habitat-sim withbullet headless -c conda-forge -c aihabitat
  

Clone the repository

git clone https://github.com/sashankmodali/LearningToNavigate.git
cd LearningToNavigate

Install Habitat-lab using following commands:

git clone  https://github.com/sashankmodali/habitat-lab.git
cd habitat-lab
pip install -r requirements.txt
python setup.py develop --all # install habitat and habitat_baselines

Datasets

Dataset can be downloaded from here
Gibson Scene datasets can be downloaded from here
Object datasets can be downloaded from here

Folder Structure

Folder structure should be as follows:

LearningToNavigate/
  habitat-lab/
  data/
    scene_datasets/
      gibson/
        Adrian.glb
        Adrian.navmesh
        ...
    datasets/
      pointnav/
        gibson/
          v1/
            train/
            val/
            ...
    object_datasets/
      banana.glb
      ...        

Create symbolic link of data folder inside Habitat Lab

cd habitat-lab
ln -s ../data data
cd ..

After setting up the environment:

1. For Milestone 1, run the following:

. milestone1.sh

OR

conda activate habitat

python main.py --print_images 1

2. To generate training data and train depth1, run the following:

. generate_training_data.sh

python train_depth1.py

OR

conda activate habitat

python main.py --print_images 1 -d ./training_data/ -el 10000 --task generate_train

python train_depth1.py

3. For Milestone 2 , run the following:

. milestone2.sh

OR

conda activate habitat

python nslam.py --split val --eval 1 --train_global 0 --train_local 0 --train_slam 0 --load_global pretrained_models/model_best.global --load_local pretrained_models/model_best.local --load_slam pretrained_models/model_best.slam -n 1 --print_images 1

python generate_video.py

4. For Final Evaluations , run the following:

. eval_ppo_st.sh

AND

. eval_ppo.sh

AND

. eval_ans.sh

Then, the results can be obtained in /tmp/dump/[experiment]/episodes/1/1/

After replacing tmp directory line in generate_video.py

To generate video, run,

python generate_video.py

Then, the results can be obtained in /tmp/dump/[experiment]/episodes/1/1video.avi