Set up the Repository#

To clone the LEAP repository, you need to send Yang your GitHub username to be invited as a collaborator.

Open terminal and go to your favorite work directory (e.g., ~/Documents) to clone the project:

git clone git@github.com:zt-yang/leap-architecture.git

Planners setup#

Add FastDownward or PyperPlan as git submodules:

cd planners
git submodule add https://github.com/aibasel/downward.git
git submodule add https://github.com/aibasel/pyperplan.git
git submodule init
git submodule update
./downward/build.py release

PDDLGym setup#

To enable our LEAP agent to interact with the world through observing and replanning (as opposted to take in fully observed state and spill out a plan), we use PDDLGym for constructing a gym-like environment (with state, step, reset) from PDDL.

cd environments
git submodule add https://github.com/tomsilver/pddlgym
git submodule init
git submodule update
brew install swi-prolog

To install swi-prolog with Homebrew, I am runing terminal from Rosetta.

PDDLGym support the following subset of PDDL1.2:

STRIPS
Typing (hierarchical is not working for Prolog inference mode)
Quantifiers (forall, exists)
Disjunctions (or)
Equality
Constants
Derived predicates (somehow a body of forall doesn't work)

It doesn't support the following extensions that our domain uses

Conditional effects
Action costs

So to integrate, I modified PDDLGym so that it ignores warning when handling those extensions.

Revised axiom due to ValueError: Cannot instantiate condition: not normalized by pddlgym/downward_translate/pddl/conditions.py

( :derived (is-egg ?i ) (forall (?i - egg) (raw ?i ) ) )

## occured only once thus i replaced predicate in precondition with the axiom body
  (:derived (has-parked ?a - agent)
    (forall (?c - car) 
      (when
        (agent-owns ?a ?c)
        (parked ?c)
      )
    )
  )

Study Downward Translator#

to test LEAP with run.py, we need to comment out the last line setup() in environments/pddlgym/downward_translate/options.py