Exploratron

Exploration is a open-source turn-by-turn sandbox puzzle game with multiple simple mechanics that interact with each other and create complex and interesting emergent gameplay situations.

Exploration is compatible with Linux, Windows and Web browser (using Webassembly).

You can play the Web browser version here.

Compilation from source

Requires Bazel.

On Web browser:

./tools/game_wasm.sh

On Linux:

./tools/game_terminal.sh

On Windows:

tools\game_terminal.bat

Map creation

Maps are created using Tiled. See examples in the exploratron\assets\map directory.

Agent training

Exploratron is also a framework to train and evaluate agent using Machine Learning.

Read, modify and and run ./tools/train_agent.sh to train and evaluate agents.

Some examples of commands (see ./tools/train_agent.sh for details):

LOGDIR=$(pwd)/training_logs

FLAG="--config=linux --define=terminal=linux_console"

# MODE="-c dbg"
MODE="-c opt"

# Run the evaluation on the "Gather" area using the Random controler (no leaning).
bazel run $MODE --config=linux --define=terminal=linux_console //exploratron/cli:evaluate_main $FLAG -- \
    --arena=Gather --controller_key=Random --num_repetitions=100

# Train and evaluate a genetic controller on the "Gather" area.
# Hyper-parameters are available in `exploratron/controller/genetic/genetic.h`
bazel run $MODE --config=linux //exploratron/controller/genetic:train_main $FLAG -- \
    --training_log_base=${LOGDIR}/gather/genetic/r1_

# Train and evaluate a hill climbing controller on the "Gather" area.
# Hyper-parameters are available in `exploratron/controller/hill_climb/hill_climb.h`
bazel run $MODE --config=linux //exploratron/controller/hill_climbing:train_main $FLAG -- \
    --training_log_base=${LOGDIR}/gather/hill_climb/r1_

# Run the evaluation on the "Gather" area using the keyboard controler (no leaning).
bazel run $MODE --config=linux //exploratron/cli:evaluate_main $FLAG -- \
    --arena=Gather --controller_key=Keyboard

Results can be compared using the R scipe ./tools/plot_training_logs.R.