vol0kin/gvgai-pddl

Implementation of a PDDL-based agent in GVGAI.

Hi there 👋

I'm Vladislav. Nice to meet you! Here you can find some information about me:

• 📐 I have a Bachelor's Degree in Computer Science and a Master's Degree in Data Science.
• 🧑‍💻 Currently I'm working as a Software Engineer developing multiplatform mobile applications.
• 🚀 I'm interested in AI in general and its applications.
• 🌱 At the moment I'm interested in learning and/or deepening my knowledge in the following subjects:
  • Tensorflow
  • PyTorch
  • 🧠 Bayesian statistics and bayesian learning
  • 🤖 Reinforcement learning

Some other facts about me:

• 🎵 I love music and I spend a lot of time listening to it. Like a lot. Oh, and I also (kinda, I'm still learning) play the guitar 🎸.
• 🧗 In my free time I like climbing (sports climbing and bouldering) and hiking.
• ⛰️ I really like mountains and nature!
• 🎮 I love video games. Also, I'm very interested in developing intelligent agents able to solve them.
• 📖 I enjoy reading books (whenever I have time for it).
• 💻 I like contributing to Open Source projects from time to time.

gvgai-pddl's Issues

Add option to save runtime information.

This should allow to save information about the generated problem files and plans.

Allow problems to be generated under a directory hierarchy.

For example:

problemFile: parent1/parent2/parent3/problem.pddl

Add debug mode.

Having a debug mode can be very useful:

• The user might understand better what the agent is doing.
• When a discrepancy is found, the decision that the agent has made can be understood better.
• When an important event happens, the user will be able to understand what is going to happen next.

Add option to remove saved goal predicates once another goal has been reached.

As the tittle says, sometimes reaching a goal means that a goal predicate that was previously reached and saved has to be removed from the list of saved goal predicates.

Add list of effects to the PDDLAction class.

As the tittle says, we must add the effects of an action as an attribute of the PDDLAction class. This will become useful when checking whether some goals are achieved.

Refactor and clean code and project.

Create discrepancy analyzer.

Add variables to save runtime information.

The following information has to be stored:

• Number of times that the planner has been called.
• Number of times that a discrepancy has been found.
• Execution time.

Automate the game information creation.

Currently, creating the data structure that represents the game information is tedious and inefficient. Thus, the creation of this data structure must be automated, so it doesn't have to be created manually.

Join JSON files into a single data structure.

The game information is currently spread across multiple files. The main idea is to join all the information in a single file, which will contain a data structure that will be loaded once a new game begins.

Test more games.

Currently, only one game has been tested. At least, 3 games have to be tested:

• Boulderdash.
• Ice and Fire.
• N/A.

Testing more games will give us a deeper insight on things that shoud be added to the data structures that we already have so that more games can be played by the agent.

Add continuos integration.

Add continuous integration to the project so that some of the modules can be tested.

Add option to call solver running on localhost.

In order not to saturate the cloud solver when some heavy testing has to be done, it could be a good idea to add an option which would allow to send the HTTP requests to a server running on localhost.

Fix repeated objects.

When creating the problem file, some of the objects are written more than once. Some planners are quite strict in this sense as they don't allow repeated objects.

Fix exception handling when the planner is called.

Currently, if the server is busy, a JSONException will be thrown. This happens while trying to create an instance of PlannerException because the code is trying to access a field that doesn't exist.

Add effects check to find out if a goal has been reached beforehand.

Add documentation to project.

Cell objects are missing when there are no associated predicates to a game object.

Even though there's no correspondence between a game object and a list of PDDL predicates, the cell objects must be generated. Otherwise, the planner will throw an error saying that there are missing objects that appear in the connectivity predicates.

Create a basic agent that has predefined goals.

This implementation must meet the following requirements:

• It has to use an agenda to store the goals.
• It has to control whether a goal has been achieved or not.
• It has to update the agenda once a goal has been achieved.

Add tests.

Add method to clear the plan instead of clearing it by retrieving it from the getter.

Allow users to pass command line arguments to the program.

Once the JAR is generated, passing arguments to it will be very handy since it will allow to modify run-time parameters (game, level, configuration file used by the agent, debug mode, etc.). Also, checking what values are passed will allow to discern whether the game will be played by an actual human or by the agent.

Move test resources to the corresponding directory.