This workshop will be an introduction to the machine learning platform developed by Unity3D called Machine Learning Agents. The Unity Machine Learning (ML) Agents platform is a great start to applying state of the art ML algorithms to games and was created to be used by everyone without a strong ML background.

The ML framework developed by Unity3D implements two unsupervised learning methods: Imitation Learning (Behavioural Cloning) and Reinforcement Learning (using the Proximal Policy Optimisation [3] ). This workshop will focus on Imitation Learning. This is used to train the agent by showing exactly how to play a game and it is a more intuitive way of training an agent than the classical trial and error approach. Attendees will also setup and configure a project in order to train their own ML agent in real-time.

Who should attend?

This workshop is targeted to beginners, without any ML prior knowledge.

Attendees should have a basic understanding of how the Unity Engine works (have worked on a least one project in Unity) and be curious and excited about how to implement imitation learning to their games.

What will happen in the workshop?

During the workshop, attendees will get a theoretical introduction to Unity ML Agents framework and the imitation learning algorithm it implements. After that, sample Unity projects will be showcased followed by an hands-on step-by-step tutorial to create a project with an ML agent/component. During the tutorial, attendees can build a game, set-up the ML-Agents framework, train it using Tensorflow and test it back in the game engine. After that, they will be able to explore more advanced features such as adding multiple agents, using offline/online learning or implementing memory-enhanced agents.

What do participants need to bring?

Unfortunately, we are not able to provide PCs/laptops for you, therefore it'll be great if you could bring your own. Also, it would be amazing (and it'll make the workshop run smoother) if you have the following installed:

Workshop Organisers: Cristina Dobre and Carlos Gonzalez Diaz

Imitation Learning for Unity Games