First, start two instances of the Malmo Client. Then run:
python pig_chase_experiment.py -t myagent
Depending on whether tensorboard is available on your system, this script will output performance statistics to either tensorboard or to console. If using tensorboard, you can plot the stored data by pointing a tensorboard instance to the results folder:
cd ai_challenge/pig_chase
tensorboard --logdir=results --port=6006
You can also run a RandomAgent or FocusedAgent baseline. Switch agents using the command line arguments:
python pig_chase_experiment.py -t random
For additional command line options, see the usage instructions: python pig_chase_experiment.py -h.
You can then navigate to http://127.0.0.1:6006 to view the results.
Our Augmented AStar agent entry for the Malmo Collaborative AI challenge is based off the AStar pathfinding algorithm. When choosing an action, it takes into account its own position and the position of the pig relative to itself, the position of the collaborator agent relative to itself and the pig, and the collaborator agent's behavior.
It also exploits the fact that the pig behaves as a normal Minecraft mob. If the pig is in an unfavorable position (eg. it can't be caught), it holds off making a move until the pig moves and goes to a position where it's easier to catch.
Below is a video of our agent playing with the FocusedAgent collaborator, as seen from a top-down view:
Below is the Reward per Episode achieved by the Augmented AStar agent:
Average score per episode: 15.53
You can evaluate our agent by running:
python pig_chase_eval.py
Results will be written to pig_chase_results.json.
Our results are also submitted to the leaderboard (team ParSys) and are as follows:
Eval Score @ 100k: 1.752 | Eval Score @ 500k: 2.113