Open 6-month internship on developing Unity-based environment for curiosity-driven deep reinforcement learning


#1

Recent developments in Deep Reinforcement Learning is at the core of the recent impressive results in Artificial Intelligence (e.g. AlphaGo). It is now possible to teach artificial agents how to play video games only from images. These kind of performances, impossible only a few years ago, are now done routinely in research labs.

The main research direction of the FLOWERS team is to study and model curiosity-driven autonomous learning, especially from the artificial intelligence perspective. In particular the team is interested in developing learning algorithms allowing agents to discover autonomously, without any supervision, how to move and perform a wide variety of skill in open ended environments.

Developing such algorithms relies heavily on the ability of researchers to be able to test their algorithms on a wide variety of environments. Until recently, researchers used to develop themselves simple, simulated environments, that would possess the desired properties. Recent progress in Deep Reinforcement Learning have made possible to tackle much more complicated environments such as open ended environments, in 3D worlds. Last year, the editor of the graphical engine Unity released a toolkit that allows researchers to test their learning algorithms on environments developed in Unity. Such environments are very interesting for they can be very rich perceptually and can contain a wide variety of interactions.

The FLOWERS team is actively looking for a software engineering intern interested in creating and developing test environments for its learning algorithms. The main part of the internship will be devoted to developing unity environments and testing curiosity algorithm in these environments. It involves both a creative part (designing rich and interesting open environments) and a technical part (implementing the environments and the software infrastructure to use them). This internship will take part in collaboration with members of the FLOWERS team working on the learning part of the project.

Candidates for this internship should ideally possess the following skills:

  • Knowledge of an object oriented programming language (Java, C#, C++, …)
  • Knowledge of a game development framework (Unity, Unreal, …)
  • Knowledge of a version-control system (Git, Svn, …)
  • Interest for game development and artificial intelligence

Period: beg. 2019 to summer 2019
Location: INRIA Bordeaux, France
Contact: pierre-yves.oudeyer@inria.fr and alexandre.pere@inria.fr

Keywords: Unity, machine learning, artificial curiosity, Deep RL, benchmark, environment, open worlds

References:




Subscribe to our RSS Feed.
Subscribe to our mailing list.