Look around and learn

Setup

• Clone this repository
• Clone submodules: git submodule update --init --recursive
• Create conda env from env.yml file: conda env create -f env.yml
• Install dependencies python -m pip install requirements.txt
• Install pkg python setup.py develop
• Move into each subdir inside third_parties and execute python setup.py develop --all
• Specify base_dir at confs/config.yaml as the absolute path of this project

Data

All the data are contained inside the data directory.

Example: habitat test scenes

1. Download test-scenes http://dl.fbaipublicfiles.com/habitat/habitat-test-scenes.zip
2. Unzip inside project directory

• Suggestion: you can keep data in a separate folder and use soft links (ln -s /path/to/dataset /path/to/project/data)

Launch an experiment

python scripts/run_exp.py run training or deployment of a policy. More information about "RL baselines" and our RL policy below.

Replay experiment

To replay an experiment, use the following python scripts/visualize_exp.py replay.episode_id={ID episode} replay.exp_name={PATH TO EPISODE} replay.modalities="['rgb', 'depth','semantic']"

Run baselines

The following learned baselines are implemented:

• neuralslam: start from confs/habitat/gibson_neuralslam.yaml
• seal-v0: start from confs/habitat/gibson_seal.yaml
• curiosity-v0: start from confs/habitat/gibson_semantic_curiosity.yaml

The following classical baselines are implemented:

• randomgoalsbaseline
• frontierbaseline-v1 (frontierbaseline-v2, frontierbaseline-v3)
• bouncebaseline
• rotatebaseline
• randombaseline

Train goalexploration policy

Start from confs/habitat/gibson_goal_exploration.yaml

• CHECKPOINT_FOLDER folder in which checkpoints are saved
• TOTAL_NUM_STEPS max number of training steps
• under ppo:
  • replanning_steps how often to run the policy
  • num_global_steps how often to train the policy
  • save_periodic how often to save a checkpoint
  • load_checkpoint_path full path to a checkpoint to load at start
  • load_checkpoint set True to load load_checkpoint_path
  • visualize if True, debug images are shown

Environments:

• SemanticDisagreement-v0 reward: sum(disagreement_t)

Environments for the RL baselines are also provided:

• SemanticCuriosity-v0 (Semantic Curiosity)
• sealenv-v0 (SEAL)
• ExpSlam-v0 (NeuralSLAM)

Policies:

• goalexplorationbaseline-v0 State: disagreement_t, map_t, agent pose

Checkpoints:

• Ours {ADD LINK}

Generate from goalexploration policy

Start from confs/habitat/gibson_goal_exploration.yaml

• replanning_steps how often to run the policy
• load_checkpoint_path full path to a checkpoint to load at start
• load_checkpoint set to True

Datasets

Scenes datasets

Tasks

Gibson object navigation (used for the main results of the paper)

You can download the task at the following link {ADD LINK}, unzip and put it in data/datasets/objectnav/gibson/v1.1

Habitat's pointnav

MatterPort

• Follow instruction in the main Habitat-lab repository

Gibson

• Ask for the license from Gibson website https://stanfordvl.github.io/iGibson/dataset.html
• Download gibson tiny with wget https://storage.googleapis.com/gibson_scenes/gibson_tiny.tar.gz
• Follow instructions at Habitat-sim to generate gibson semantic

Dependencies

• detectron2 >= 0.5
• torch >= 1.9
• pytorch-lightning >= 1.5
• habitat-sim = 0.2
• habitat-lab
• torchmetrics >= 0.6

Contributing

If you want to contribute to the project, I suggest to install requirements-dev.txt and abilitate pre-commit

python -m pip install -r requirements-dev.txt
pre-commit install