Setting up project
- Ubuntu 18.04+
- Python 3.8+
Check out instructions to use the Dockerfile if you don't meet these requirements
Clone the repo:
git clone [email protected]:zkytony/cos-pomdp.git
After cloning do the following three commands separately.
source setup.bash
source setup.bash -I
source setup.bash -s
It is recommended to use a virtualenv with Python3.8+
To test if it is working: Do pip install pytest and then go to tests/ folder and run
pytest
You are expected to see something like:
-- Docs: https://docs.pytest.org/en/stable/warnings.html
======= 14 passed, 2 skipped, 3 warnings in 43.71s ======
Note: setup for SAVN, MJOLNIR etc. were attempted during the project; MJOLNIR can run but does not work well.
At this point, you should be able to run a basic object search domain with COS-POMDP. A
python -m cospomdp_apps.basic.search
pygame window will be displayed
(Skip this if you are running on a computer connected to a display) If you are running offline, or on a server, make sure there is an x server running. You can do this by:
sudo Xorg -noreset +extension GLX +extension RANDR +extension RENDER -config xorg_conf :0
Note that depending on your local configuration you may want to use something other than :0. If there is already an x server running at :0 but your $DISPLAY shows nothing, you should run it at another display number.
Now, download necessary models by running the following script. This will download three files: yolov5-training-data.zip, yolov5-detectors.zip, corrs.zip, place them in the desired location and decompress them.
# run at repository root
python download.py
Then, main test we will run is
cd tests/thor
python test_vision_detector_search.py
This will run a search trial for AlarmClock in a bedroom scene. When everything works, you may see something like this:
Note that the search process may vary due to random sampling during planning.