This project aims at designing a popular science experiment in the form of an in-browser interactive demonstration in order to showcase the capabilities of Deep Reinforcement Learning agents, including their ability to generalize to unknown tasks. The demonstration is available online here.
The repository is organized as follows:
📦 Interactive_DeepRL_Demo
┣ 📂 policy_models -- policies available for each morphology in the TF SavedModel format
┣ 📂 web_demo -- web app main folder
┃ ┣ 📂 base_envs_set -- set of basic environments in JSON files
┃ ┣ 📂 images -- graphics used in the demo
┃ ┣ 📂 js -- source code
┃ ┃ ┣ 📂 bodies -- code for agents morphologies
┃ ┃ ┣ 📂 Box2D_dynamics -- collisions handlers
┃ ┃ ┣ 📂 CPPN -- weights and code of the CPPN used to generate terrain
┃ ┃ ┣ 📂 envs -- available environments
┃ ┃ ┣ 📂 ui_state -- UI state management
┃ ┃ ┣ 📂 utils -- utility classes and functions
┃ ┃ ┣ 📜 box2d.js -- full box2d code
┃ ┃ ┣ 📜 draw_p5.js -- rendering functions
┃ ┃ ┣ 📜 game.js -- handles simulation execution
┃ ┃ ┗ 📜 i18n.js -- handles internationalization
┃ ┣ 📜 index.html
┃ ┣ 📜 index.js
┃ ┣ 📜 ui.js -- sets up the different UI elements
┃ ┗ 📜 demo.css
┣ 📜 list_base_envs.py -- python script used to generate a JSON file which lists all the files in base_envs_set
┗ 📜 policies_to_json.py -- python script used to generate a JSON file which lists all the policies in policy_models
Follow these steps if you want to launch the demo locally.
1. Get the repository
git clone https://github.com/flowersteam/Interactive_DeepRL_Demo.git
cd Interactive_DeepRL_Demo/
2. Install it, using Conda for example (use Python >= 3.6)
conda create --name DRLdemo python=3.6
conda activate DRLdemo
pip install tensorflowjs
3. Set it up
3.1. Convert all policy models in policy_models to a web-friendly format in web_demo/policy_models
ls -d policy_models/*/*/*/ | xargs -I"{}" tensorflowjs_converter --input_format=tf_saved_model --output_node_names='parkour_walker' --saved_model_tags=serve --skip_op_check {}tf1_save web_demo/{}
3.2. Generate the list of policy models
python3 policies_to_json.py
3.3. Generate the list of files in base_envs_set
python3 list_base_envs.py
4. Launch the web app
pushd web_demo/; python3 -m http.server 9999; popd;
If you use our demo in your work, use the following citation:
@misc{germon2021demo,
title={Interactive Deep Reinforcement Learning Demo},
author={Germon, Paul and Romac, Clément and Portelas, Rémy and Pierre-Yves, Oudeyer},
url={https://developmentalsystems.org/Interactive_DeepRL_Demo/},
year={2021}
}