Implementation of LAO*/ILAO* [Hansen and Zilberstein, 2001] algorithms to solve MDPs described as PDDLGym environments
This implementation uses the function get_successor_states imported from PDDLGym's core module (pddlgym.core).
Since this feature is currently available in PDDLGym's repository but not in its latest pypi release as of now,
to use it you'll need to either clone the repository and install it locally or install it via pip by pointing to the repository.
You can do the former by settting up a virtual env (see here) or the latter by running the following:
$ pip install git+https://github.com/tomsilver/pddlgym
For usage instructions, run python main.py --help in the repository's root folder
The following command example can be used to solve the first available instance of the Triangle Tireworld [Little et al., 2007] environment using the ILAO algorithm and output some results:
$ python src/main.py --env PDDLEnvTireworld-v0 --problem_index 0 --algorithm ilao --render_and_save
To simulate an episode after the optimal policy has been found and render each encountered state:
$ python src/main.py --env PDDLEnvTireworld-v0 --problem_index 0 --algorithm ilao --simulate --render_and_save
Right now only the heuristic function h(s) = 0 is available, so the performance might not be great.