/Unity-ML-Agents

A small project exploring the basic premise of reinforcement learning using Unity's ML agents. The project trains a game object to jump over obstacles at the correct time to score a point.

Primary LanguageC#

Unity ML Agents

A small project exploring the basic premise of reinforcement learning using Unity's ML agents. The project trains a game object to jump over obstacles at the correct time to score a point.

Commands

Installing ml-agents

Create a Virtual Environment and install mlagents. Python is a prerequisite.

  1. cd into an empty folder where you want to create the virtual environment 
    py -m venv ExampleNameEnv
  2. To activate Virtual Environment, cd into virtual environment location and 
    .\Scripts\Activate
  3. While the virtual environment is activated, 
    pip install mlagents

Training

  1. cd into yaml location inside the unity project 
    mlagents-learn Jumper.yaml --run-id="JumperTestx"
  2. Training a build.exe 
    mlagents-learn Jumper.yaml --run-id=JumperTestx --env= ./Build/Unity-ML-Agents.app --time-scale=10 --quality-level=0 --width=512 --height=512

Tensorboard Logs

  1. cd into location where "results" are stored 
    tensorboard --logdir=results

Results

Environment

Losses

Policies

Key

Test Scenes

Scene 1

Scene 2

Built With