Technologies : Python, Tensorflow

This was a fun one week project to get a better understanding of the PPO algorithm

Using as a reference the ppo paper, David Silver's course on RL and openAI baselines I reimplemented the ppo algorithm and tested it on two gym environments: CartPole and Atari Breakout.

I ended up getting similar results to the openAI ppo baseline on these two environments.

Afterwards, I also added curiosity based rewards based on this work and tested it on an environment where an agent must click buttons appearing at random positions

Technologies : Python, Tensorflow, Unity ML agents

The objective of this project was to teach agents to play my game Slideball.

I got the agents to pick up a ball and throw it into goals placed at random positions. And later started to make them play against each other. For that I had to add new features to the ML Agents framework to allow agents to play againt past versions of themselves and an elo rating system to be able to measure the progression of the agents skill level.

You can see the agents training in this video

Technologies : Unity with C#

This is a puzzle game that progressively introduce new mechanics to the player that they must understand to progress further. At the moment there are only puzzles and no real structure to navigate them.

Implementing the undo feature while having animations between the different board states was an interesting problem. That was one of my first projects and I was excited to use all the C# features I learned about. So event-based architecture makes things way more complicated than they should be :P

This game is multiplayer, it is better to play with friends.

Technologies : Unity with C#

Tools : GIMP, Blender, Maya, Audacity

I really liked the mod Hover on Starcraft II and built a browser version to learn about networking in games.

It uses WebRTC to play peer to peer with friends and can be pretty fun to play 2v2 or 3v3. After discovering Reinforcement Learning in 2018 I started modifying it to teach agents to play the game

Technologies : Unity with C#, XML
Tools : GIMP

Working with 3 other students, we made a game to teach aeronautics to children from 6 to 10 years old and produced a small paper containing recommandation for creating educational entertainment.

After learning how to use Unity, I created two of the 9 small games, the save management and translation modules and half of the game UI and navigation. I ported the game from android to WebGL.

I also organised and conducted user tests with primary school children.