Overview / Usage

A fun game utilizing reinforcement learning.

A crawler with 4 arms is trained to move in the direction of the target. The crawler receives a reward for facing the target, and a reward for stepping in the direction of the target.

Methodology / Approach

Leverages research on continuous control and locomotion from OpenAI and DeepMind using Unity3D.
The Unity allows the simulation and control one or more of the agents in the Scene via a
keyboard or game controller. This can be used for generating expert demonstrations to train
imitation learning algorithms.

The ML-Agents toolkit is an open source project which enables researchers and developers to create
simulation environments using the Unity Editor and interact with them using a Python API.

Tensor Flow Sharp would need to be activated as a wrapper for tensor flow library.

Technologies Used

ML-Agents Library
Tensorflow Sharp
Unity3d
Visual Studio

Repository

https://github.com/Spencerish/symmetrical-potato