Reinforcement Learning exhibit for the I AM A.I. exhibition (v2)
To install the required dependencies run npm install
in the root directory.
You can use npm run build
or npm run watch
in the root directory to build the client apps.
The .env
file in the root directory contains settings that are applied at compilation time.
The config directory has several data definitions.
You can override any of them by creating a settings.yml
file in the root directory.
exhibit.html
runs the exhibit in a fixed 1920x1080 resolution.
To override settings in exhibit mode create a settings-exhibit.yml
file in the root directory.
The default language can also be set through the lang query string (e.g. ?lang=de
).
embed.html
allows you to embed instances of the app in your own website via an iframe.
You can customize the functionality of the embeded app through query strings arguments:
- lang (default:
en
): Language. - map (default:
maze1
): Map to show. Maps have to be stored indata/mazes
. - training: Name of the q-value table to initialize the robot with. Q-values are stored in
data/training
. - editmap (default:
false
): If true, the map can be edited. - tiles: Comma separated list of the ids (numerical) of the tiles to show in the left side palette. Ids can be viewed in
config/tiles.yml
. - cmds: Comma separated list of UI elements to show, from:
- run: The run button
- step: The step button
- turbo: The turbo button
- clear: The button to clear the training
- xr: The exploration rate slider (Explore / Exploit)
- reset-map: The button to reset the map
- policy: The button to show the policy and value function
- xr (default:
0.2
) The starting exploration rate. - lr (default:
1
) The learning rate. - speed (default:
10
): The speed of the robot. - showpolicy (default:
false
): If true, the policy and value function are shown. - showqv (default:
false
): If true, the q-values are shown on the map. - autorun (default:
false
): If true, the robot starts in the running state. - showrewardbar (default:
false
): If true, the reward bar is shown over the map (currently very specific to a particular experiment, so probably not practical).
Developed by Eric Londaits for IMAGINARY gGmbH adapted from the original reinforcement-learning exhibit by Sebastián Uribe and Andreas Matt.
Dutch translation by Jarne Renders. Spanish and French translations by Daniel Ramos.
Copyright (c) 2020-2021 IMAGINARY gGmbH Licensed under the MIT license (see LICENSE)