AUTONOMO lets the player create and train a reinforcement learning agent, no programming or machine learning knowledge required. The player can not only create the agents, but also the environment in which they train.
Date: 6th Semester (2018)
Time limit: 4 months
Team size: 2 People
Engine: Unity 3D + Python
Constraint: Develop a game based on Unity’s ML-Agents Toolkit (v0.3)
My Part: My main task was to extends Unity’s ML-Agents toolkit, to make it possible to train, save and load agents in a build. I also implemented the logic for serialization of the levels.
What I learned: I learned how reinforcement and imitation learning can be used to train agents in the context of the ML-Agents toolkit. In addition I figured out how serialize nested data structures that contain children with references to each other in the context of Unity. Furthermore I acquired the skills necessary to implement inter process communication using sockets between Unity and Python.