Here we will be posting experiments that never turned into successful projects but you might find interesting.
Do you think there is potential in any of these projects? Let us know!
A tiny match-3 experiment where you steer balloons in real
time by controlling fans on both sides of the screen. It's one of our
early small games.
An early real-time puzzle game where you build a path for a
chicken to reach its egg. It remains available as a small historical
project, but no longer belongs with the studio's adventure-game catalog.
Quaderno is a tool to create adventure games with very little
programming. You create your adventures by writing an HTML file where you
declare the rooms and interactions, and the quaderno engine does the rest.
It's a great companion if you want to test quickly your ideas before
spending time coding animations or making complex graphics.
I thought it would be fun to create a version of the breakout
classic using a neural network. Note that the neural network is not playing
the game. The neural network _is_ the game (the input to the network are
player moves, and the output is the rendered image to put on the screen). In
the GIF at the left, at the top you have the game programmed in JS, and at
the bottom the game programmed as a neural network, both rendering the same
gameplay. If you are into machine learning, the model uses 4 layers of LSTM
networks, with 100 cells on each layer.