This is the graphical simulation developed in Unity of the real world experiment from Shafti et al. (2020) [1] used from the MazeRL
Forked from https://github.com/panos-stavrianos/MazeUnity and adjusted accordingly for MazeRL
The environment receives actions (plus other important information) and sends back observations (plus other important information) to MazeRL.
The above messages are being exchanged via an HTTP server.
MazeRL and MazeUnity work as HTTP clients.
-
Download git
git clone https://github.com/ligerfotis/MazeUnity.git -
Install Unity (Version: 2020.3.13f1)
-
Open with unity
- If Unity cannot open the project or does not recognize it, create a new project and replace the folders.
-
Remove default scene
-
Drag and drop main scene from Scenes in the Hierarchy
https://www.jetbrains.com/help/rider/Unity.html#getting-started
- Start the dedicated Maze Server
- Start the experiment MazeRL
Just open the game in Unity and press the play button.
See here for connection instructions
Every time a user opens the link to the webgl in the browser the game is being sent to it from a docker.
-
Build Settings -> web_build -> (switch platform) -> build
-
Choose web_build and name it “webgl”
-
Go to to MazeUnity/web_build
cd MazeUnity/web_build -
Edit webgl.conf 'server_name' to the name of your server without the ‘http://’ header (default: localhost)
-
Make sure the ports used below are open.
-
docker build -t <image_name>:<version> . (docker build -t maze-unity:1.0.0 .) -
docker run -p <host_port>:80 <image_name>:<image_version> (docker run -p 12000:80 maze-unity:1.0.0) -
Open <server_name>:<host_port> (localhost:12000) in a browser.
-
If you want to stop the docker:
docker stop maze -
If you want to remove the docker:
docker rm maze
In Assets/Scripts folder.
- Tray: 50cm x 50cm
- Wall thickness: 3cm
- Obstacles opening: 9 cm
- Ball radius: 5cm
- Hole radius: 5 cm
[1] Shafti, Ali, et al. "Real-world human-robot collaborative reinforcement learning." arXiv preprint arXiv:2003.01156 (2020).