Code for the paper Hierarchical Representation and Explicit Memory: Learning Effective Navigation Polices on 3D Scene Graphs using Graph Neural Networks published at the International Conference on Robotics and Automation (ICRA) 2022.
We reccomend installing in a virtual environment, such as Anaconda or venv. We use Python 3.7 in our work.
Clone and install this repo:
git clone git@github.mit.edu:TESS/dsg-rl.git
cd dsg-rl
python -m pip install .
Next, install Pytorch geometric.
Pytorch geometric is sensitive to PyTorch and CUDA versions, so we recommend that you install this dependency by following the author's instructions. You can check your PyTorch and CUDA versions via the following command
>> python
>> import torch
>> torch.__version__
1.10
>> torch.version.cuda
11.3For example, installing via pip with PyTorch version 1.10 and CUDA 11.3 would look like:
pip install torch-scatter torch-sparse torch-cluster torch-spline-conv torch-geometric -f https://data.pyg.org/whl/torch-1.10.0+cu113.html
See the provided train-template.yaml for an example configuration file. Note that any fields in will need to be updated with the correct device-specific path. In particualr, paths to the TESSE simulator, Dynamic Scene Graphs (DSGs), and Euclidean Signed Distance Functions (EDSFs) must be provided, all of which may be found here.
The train-dsg-rl.py script may be used to train an agent.
>>> python ./scripts/train-dsg-rl.py -h
usage: train-dsg-rl.py [-h] [--config CONFIG] [--name NAME]
[--timesteps TIMESTEPS] [--sim-ok] [--restore RESTORE]
optional arguments:
-h, --help show this help message and exit
--config CONFIG Training configuration file.
--name NAME Experiment log directory name
--timesteps TIMESTEPS
Number of simulation timesteps for which to train.
--sim-ok Start training even if there are other simulator
instances running. If true, be sure that simulation
ports don't conflict.
--restore RESTORE Restore training from this checkpoint
The eval-dsg-rl.py may be used to evaluate and agent.
>> python ./scripts/eval-dsg-rl.py -h
usage: eval-dsg-rl.py [-h] [--config CONFIG] [--ckpt CKPT]
[--episodes EPISODES] [--sim-ok] [--log-path LOG_PATH]
optional arguments:
-h, --help show this help message and exit
--config CONFIG Agent configuration
--ckpt CKPT Checkpoint to evaluate.
--episodes EPISODES Number of episodes to run
--sim-ok Start training even if there are other simulator
instances running. If true, be sure that simulation
ports don't conflict.
--log-path LOG_PATH Name of evaluation result directory.
DISTRIBUTION STATEMENT A. Approved for public release. Distribution is unlimited.
This material is based upon work supported by the Under Secretary of Defense for Research and Engineering under Air Force Contract No. FA8702-15-D-0001. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Under Secretary of Defense for Research and Engineering.
(c) 2020 Massachusetts Institute of Technology.
MIT Proprietary, Subject to FAR52.227-11 Patent Rights - Ownership by the contractor (May 2014)
The software/firmware is provided to you on an As-Is basis
Delivered to the U.S. Government with Unlimited Rights, as defined in DFARS Part 252.227-7013 or 7014 (Feb 2014). Notwithstanding any copyright notice, U.S. Government rights in this work are defined by DFARS 252.227-7013 or DFARS 252.227-7014 as detailed above. Use of this work other than as specifically authorized by the U.S. Government may violate any copyrights that exist in this work.