AUV Simulation in Unity using ML-Agents

Trained using Deep Reinforcement Learning

Dependencies

1. Unity 2019.3.15f
2. Python 3.6+
3. ML-Agents release_2
4. mlagents - python package v0.16.1

Install

Refer full guide here

Basically:

1. Clone the ML-Agents Toolkit Repository
2. Install the com.unity.ml-agents Unity package from the package manager (ensure you enable preview packages & download 1.0.2)
3. Install the mlagents Python package v0.16.1 exactly

Run and Train

1. For running trained model, just click play. The main scene is Scenes/Arena
   • The AUV Gameobject has an agents script which has the neural network stored
   • Neural Network trained models are in /Assets/TrainedModels/models
2. For training, open terminal in /Assets/TrainedModels/models
   • Run

   mlagents-learn auv_config.yaml --run-id=runID

   • auv_config.yaml has the reinforcement learning training parameters
   • The model will be saved in the models directory
3. For visualizing the data on tensorboard: release_2_docs
   • Run

   tensorboard --logdir=summaries

   • Navigate to http://localhost:6006/ for the results

To-do

1. Train the boxed version

Learn

Note: only refer to the correct version of ML-agents documentation.

• Excellent starting guide
• Example Environments
• Creating a new Environment and Agent