Generate physical reasoning datasets with Box2D.

The simplest way to install PHYRE is via pip. As PHYRE requires Python version 3.6, we recommend installing PHYRE inside a virtual environment, e.g. using Conda.

Notice that currently you can only install PHYRE with pip in Linux. See this issue for latest updates on Mac.`

conda create -n phyre python=3.6 && conda activate phyre

pip install phyre

sudo apt-get install -y build-essential python-dev swig python-pygame git

git clone https://github.com/pybox2d/pybox2d

cd pybox2d

python setup.py build

sudo python setup.py install

Tasks in PHYRE are defined with template id and mod id, e.g., task with id 00000:001 has 00000 as its template id and 001 as its mod id. You can either do a specific task with --task_id or a range of tasks with --start_template_id, --end_template_id and --num_mods.

To run the program, cd into GenBox2D/src/main/python and run main.py. For example:

# Single task
python main.py --task_id 00000:001

# Range tasks
python main.py --start_template_id 0 --end_template_id 5 --num_mods 2

There are two modes in GenBox2D, GUI mode and dataset mode

GUI mode allows you to visualize the world created by pyBox2D with pygame package for graphics. To enable GUI mode, you need to set -i flag when running the main function. GUI mode will show the graphics without generating datasets. You can use keyboard to switch between tasks and use mouse to drag around the objects.

An example command to run the program in GUI mode:

python main.py --start_template_id 0 --end_template_id 5 --num_mods 2 -i

Some controls when interacting with GUI:

• Enter: switch to the next task.
• Backspace: switch to the previous task.
• ESC: exit GUI.
• Arrow keys: Move the screen.
• Scroll wheel: Zoom the screen.

Dataset mode will generate datasets with the way you want. The generated data are organized in the following file structure.

log-06-17-2020-01-15-31
├── 00000:000.log
├── 00000:001.log
├── 00001:000.log
├── 00001:001.log
├── 00002:007.log
├── 00002:011.log
└── ...

For each task, we save the information for the task, the action, simulation results for every timestamp, and the goal result into JSON format. You can see the example JSON files under the GenBox2D/example folder. Each log file has 4 types of JSON, the first line is for the task infomation, the second line is for the action information, from the third line each line has the information of objects in one timestamp, and the last line is the JSON shows the task is solved or not.

GenBox2D provides multiple parameters to customize and prepare your datasets. Simply run python main.py -h to see these parameters. Also, you may want to change the physical properties by creating another config JSON under the GenBox2D/config folder.

Run python replay.py [log_path] to visualize the dataset, generating images or gifs from the log file. some configurable parameters:

• --log_path: the path of the log file. It can be either a single file or a directory (will iterate through all log files under the directory).
• --dir: use this flag to indicate the log_file is a directory
• --image: generate jpg image for each frame
• --gif: generate gif